[HN Gopher] Qdrant, the Vector Search Database, raised $28M in a...
___________________________________________________________________
Qdrant, the Vector Search Database, raised $28M in a Series A round
Author : francoismassot
Score : 122 points
Date : 2024-01-23 10:34 UTC (12 hours ago)
(HTM) web link (qdrant.tech)
(TXT) w3m dump (qdrant.tech)
| redwood wrote:
| Anyone using Qdrant in prod?
| andre-z wrote:
| Many: https://testimonial.to/qdrant/all
|
| https://techcrunch.com/2024/01/23/qdrant-open-source-vector-...
| redwood wrote:
| I should have been clearer in my question: It would be great
| to hear directly from people who are using them about their
| successes and what their experience has been like
| andre-z wrote:
| Does this count? https://testimonial.to/qdrant/all
| swalsh wrote:
| I built a little proof of concept that uses it in the RAG
| pipeline, it's been proving quite useful, so we're just
| starting the move to production.
|
| It's probably going to stay, but I'm also evaluating databricks
| new vector store as we're using databricks for all the
| analytics parts of the app already, and having them all on the
| same infrastructureis appealing.
| crucio wrote:
| We are for a few projects. We've been using them for over a
| year and have been impressed. We have 10s millions of items in
| there with lots of daily inserts/deletions etc. There's been a
| couple of gotchas but generally it is quite predictable and
| scalable.
|
| We use 768 dimensional vectors for our items with several other
| payload filters (e.g. language). Performance has been good and
| I think the qdrant team focus on the right features without
| creeping into other areas.
| inertiatic wrote:
| I do, and it's very very rough around the edges to be honest.
| Lots of things broken, things are even breaking between
| releases suddenly in unexpected places. Or at least, I'm used
| to working with more robust data stores. If my work was more
| high stakes, I'd have already advocated for moving our vector
| search to something more robust. Thankfully it's not and I can
| just maintain what we're making with not too much stress, and
| enjoy seeing this OS project grow from a user perspective
| (haven't seen a data store go through this very initial phase
| in my career yet).
|
| Support from the team is great however, and congrats to them
| for this round!
| esafak wrote:
| Please elaborate. What would you have moved to, for example?
| This is valuable information.
| clbrmbr wrote:
| What's the mote here? Seems to be a risky investment when it's
| such a crowded space and likely to be decent open source
| alternatives for those with small budgets and homegrown solutions
| for companies with bigger budgets and requirements.
| jjackson5324 wrote:
| It's a series A. Thinking about the moat in a hot market like
| vector DBs is a great way to miss out on unicorns.
| tw1984 wrote:
| the OP argued that it is not a hot market - companies like
| openai is going to eventually use its own while small players
| are going to just use openai's assistant APIs, they don't
| have to operate their own "vector database".
|
| it is also worth to mention that even if there is going to be
| a market called "vector databases", which is highly unlikely,
| you can't just written off all existing regular databases and
| pretend that they are not going to just walk in and take
| over.
|
| all in all, there is no reason to believe it is a hot market.
| it is much better to ask is there going to be a market at
| all.
| mritchie712 wrote:
| OpenAI uses Qdrant
|
| https://news.ycombinator.com/context?id=38611608
| tw1984 wrote:
| for now.
|
| let me repeat what I have already explained - when
| compared to today's leading AI tech, a "vector database"
| is just ancient tech. major players are going to build
| their inhouse solution or they'd conclude it to be some
| kind of labor intensive & low profit margin baggage and
| outsource it.
|
| you can build a business around it, just like all major
| tech companies have cleaning guys work for them one way
| or another, people have to realize that it doesn't make
| carpet cleaning a high tech or strategically important
| business.
| nostrebored wrote:
| Work into performant vector search is an active area of
| research. If it were such a commodity, there wouldn't be
| such a wide variance in performance among existing
| solutions.
|
| There are a ton of open questions. If you think about
| Elasticsearch as a similar domain, you have complexity at
| the ingest, storage, and horizontal scalability layer. If
| you think places are going to invest in their own
| distributed system that handles these components, I think
| you'd be as wrong as saying that people will invest in
| their own managed Lucene implementations.
| nostrebored wrote:
| For a comparison of existing assistants API vs. vector
| search, you can check out my blog at
| https://nostrebored.com
|
| At a high level there are a few differences
|
| - Control over embeddings. What gets embedded? What are the
| output vectors? What models do you use? How do you handle
| multimodal input?
|
| - Performance. When you make a call to Assistants, you have
| to wait for the Assistant to understand that it needs to do
| RAG. This performance hit is actually quite large (look at
| the two videos on the blog for reference)
|
| - Cost. OpenAI has an incentive to load the context window
| to consume more tokens. A few dozen calls to Assistants was
| costing me around $10.
| lyjackal wrote:
| Qdrant is open source (with hosted offerings).
| mindvirus wrote:
| Congrats to them!
|
| What have your experiences with vector databases been? I've been
| using https://weaviate.io/ which works great, but just for little
| tech demos, so I'm not really sure how to compare one versus
| another or even what to look for really.
| danielbln wrote:
| We're using Postgres with the pg_vector extension for basically
| all of our projects. We know and love Postgres, it has a big
| track record, the extension is supported on all major managed
| cloud offerings, no new tooling needed, pg_vector supports HNSW
| indices for performance as well.
|
| Once in a whole supabase slips into a project, but that's
| basically just Postgres with some bells and whistles on top.
|
| I got nothing bad to say about Pinecone, We aviate, Chroma etc.
| but when it comes to dbs, I like to go with the devil I know.
| andre-z wrote:
| You should use whatever works best for you unless you face
| some limitations. The issue is that vector databases are not
| databases but search engines. It is ACID vs BASE. A few
| thoughts on this https://qdrant.tech/articles/dedicated-
| service/
| Fendyfd wrote:
| There are multiple vectordb available in the market, open
| source ones include Milvus, Qdrant, Weaviate etc. Cloud
| services include Zilliz Cloud (managed milvus), Qdrant Cloud,
| Weaviate Cloud etc. Try using a benchmark tool to evaluate
| them. Here is an open-source option for your reference:
| VectorDBBench (https://github.com/zilliztech/VectorDBBench)
| tw1984 wrote:
| why would anyone use a "benchmark" tool from a vendor (zilliz
| here) to test the performance of its competitors?
| swalsh wrote:
| I've been using Qdrant. Can't speak highly enough of the core
| functionality. It's fast, good accuracy, easy to use etc.
|
| I think there are some things I wish were easier, for example
| finding and updating points, and the UI could be better.
| bootsmann wrote:
| I think one of the big advantages of qdrant is how easy it is
| to do a poc because it allows you to have an "in-memory"
| version of the database similar to sqlite. One of the big
| competitors, Milvus, comes with a fairly intricate docker-
| compose you have to spin up to try it.
| J_Shelby_J wrote:
| Only for python!
| avereveard wrote:
| Well sqlite now has a vector extension so it's super
| convenient for testing. Between that and fts5 sqlite can
| stand in for any advanced search service as far as poc are
| concerned.
| weinzierl wrote:
| A while ago I read in a thread here that they are used in
| OpenAI's products and at another popular company. I am not sure
| but vaguely remember X/Grok.
|
| They are also a Rust shop.
|
| Who says Germany has no cool startups.
|
| EDIT: Yes, it was Grok.
| sroecker wrote:
| Yes, used by Grok:
| https://twitter.com/qdrant_engine/status/1721097971830260030 Oh
| wow, completely missed that they're German. Should have noticed
| their "Impressum"..
| treprinum wrote:
| Is Open AI using it in their assistants API for retrieval?
| Answer performance of those is really bad and retrieval is slow
| compared to Pinecone.
| simonw wrote:
| Yes, it's used for the RAG implementation - though we only
| know that due to information leaked in an error message I
| believe:
| https://twitter.com/altryne/status/1721989500291989585
| saliagato wrote:
| Hard to believe OpenAI uses Quadrant when they are backed
| by Microsoft, thus having Azure Cognitive Search (now "AI"
| Search)
| wodenokoto wrote:
| Do you have any experience in AI search to compare it to
| other products?
|
| I'm genuinely curious to know if it's any good.
| chintler wrote:
| Cognitive Search is nowhere as good as a 'pure' vector
| DB. Behind the scenes, it's a managed
| elasticsearch/opensearch with some vector search
| capabilities. The 'AI' implementations I've done with
| Cognitive Search always boil down to hybrid(vector+fts)
| text search.
| treprinum wrote:
| Cognitive Search already contains hybrid search (vector +
| BM25 + custom ML reranking) and they use chunks of 2048
| tokens with a custom tokenizer. So it should be now
| better than most vector DBs. One could probably make
| something better by using some version of SPLADE instead
| of BM25 but their secret sauce lies in their custom ML
| model for reranking that gives them the largest search
| performance boost.
| lazydon wrote:
| In context of RAG, the goal is not to have a pure vector
| DB but to have all the relevant data that we can gather
| for a user's prompt. This is where Cognitive Search and
| other existing DBs shine because they offer a combination
| of search strategies. Hybrid search on Cognitive Search
| performs both full text and vector queries in parallel
| and merges results which I find a better approach.
| Further, MS is rebranding Cognitive Search as Azure AI
| Search to bring it more in line with the overall Azure AI
| stack including Azure OpenAI.
| infecto wrote:
| Simon, I am always amazed how you are able to keep up with
| so much so quickly!!
| shanghaikid wrote:
| Congratulations.
|
| What do you think you milvus? https://milvus.io/. The difference
| seems significant from the architecture perspective.
| prakhar897 wrote:
| I applied to Qdrant a while back and got this response:
|
| "We are getting many applications for this position. Usually, a
| test task would help preselect suitable candidates. However,
| since we develop open-source software, we rely on contribution.
|
| You can build an open-source Qdrant connector to another
| framework or library. The simplest one would be, for example, a
| Streamlit data connector. But other ideas are more than welcome!
|
| No limitations and no deadline. As long as this job position is
| online, we accept submissions. After you are done, send us an
| email to career@qdrant.com with the link to the repo. We will
| review it and get back to you asap."
|
| No interviews, conversation before this email. Hope they see and
| fix this.
|
| Edit : No Pay.
| Oras wrote:
| Do you expect a tech interview by just applying?
|
| From your perspective, you're filling out an application, maybe
| writing a cover letter, but on the other side, there are 100+
| applications like yours. Not all of them are qualified, CVs are
| not a trustable source anyway.
|
| That's why companies add tests to filter first, then interview
| later.
| prakhar897 wrote:
| I don't expect interviews. But I also don't want to spend 20
| hours working before getting a "Unfortunately we've decided
| not to move forward" message.
|
| As a thumb rule, I'm happy to put 4x more effort than the
| company. If they interview me for 1 hour, I spend 4 hours
| doing the take-home. Anything more feels like exploitation.
| epolanski wrote:
| Well, they said they receive a lot of applications so they
| are in the position of setting the rules.
|
| You are absolutely right into setting your own rules as
| well, those haven't overlapped.
| pclmulqdq wrote:
| As a general rule of thumb, random series A startups are
| in much lower demand for top-tier talent than top-tier
| talent is in demand for these companies. That would mean
| that the good engineers should set the rules of
| engagement, and that any startup that thinks they set the
| rules is attracting worse talent.
| Galanwe wrote:
| Well, unless Qdrant writes a post complaining about the
| quality of their applicants, I don't see where the issue
| is.
|
| Also, not all companies try to maximize for "top tier
| developer", it seems they are maximizing for "top
| motivated developer", which does not seem stupid either.
| andre-z wrote:
| Thanks. We will reveal our hiring strategy in more
| details soon.
| guappa wrote:
| Maybe they maximize for "top gullible young inexperienced
| person that will fall for that"?
| pclmulqdq wrote:
| It sounds like they are instead maximizing for "free
| integrations," which seems to be a fine way to get
| neither free integrations nor high-quality candidates.
| epolanski wrote:
| The general rule of thumb is that companies are free to
| decide what they look for and how to find it and so are
| candidates.
|
| This is an open source database company, obviously the
| best candidates are contributors to the project first.
|
| Picky ivy league graduate that farms leetcode and system
| design question is very very low in the ranking of what
| such companies should look for.
| pclmulqdq wrote:
| All the ivy league graduate leetcode farmers I know are
| actually the ones who _would_ do the grunt work of
| developing database integrations for free if they
| believed a decent salary at a prestigious job might be
| waiting over the hill.
|
| The people I have met with the lowest tolerance for this
| stuff are the ones who actually produce the most
| impactful work. Partly because they don't do work that
| has no impact on their lives.
|
| Edit: Obviously, they can do whatever they want, but that
| doesn't mean that it's a good sign from outside.
| treprinum wrote:
| "We receive a lot of applications" can be also a
| marketing speech and it could also mean they are flooded
| by spam requests from all over the world they can't
| filter out.
| otabdeveloper4 wrote:
| Then this job is not for you, simple as. Sounds like a win-
| win.
| 0xedd wrote:
| Don't be ridiculous.
|
| HR gets paid to talk to candidates. I don't get paid to
| apply. The initial screening call is what allows a company to
| gauge the relevancy of a candidate. Let him speak about some
| of the topics and see how in-depth they go. Either the HR is
| familiar enough with the tech to understand proficiency
| (think a student listening to a maths professor) or they let
| a TL have a short conversation. I've overheard unqualified HR
| do their jobs badly, too; They laughed at picking them by
| looks and "feels". But, that's out of scope here.
|
| A large company has millions to invest in different areas.
| Intrinsically, it has a much larger margin of error. You
| accidentally overprovisioned some resources and cost the
| company 10k? Tis but a scratch. You POC some personal project
| and accidentally get billed 10k? That is not the same.
|
| A company can spend money on hiring. It is expected to. A
| private person can't spend money on applying to jobs. It
| isn't expected.
|
| It's interesting to see how the shift goes from the self to
| the company [and to the country]. A little bit of communist
| propaganda goes a long way, eh, comrade NPC?
| unglaublich wrote:
| Companies not paying for their applicant's time is
| communist now?
| guappa wrote:
| > NPC
|
| When someone calls others "npc", I understand that they are
| complete psychopaths that are somehow capable of thinking
| that the other people don't live the full human experience
| as they do.
| redwood wrote:
| That's a pretty smart way for them to seed the ecosystem with
| open source connectors! Are you implying that that's what they
| were really trying to do here? Or do you think it was a genuine
| filter technique?
| simonw wrote:
| If they're paying you for your time, this is kind of smart.
|
| If they're not then it's a scandal.
| prakhar897 wrote:
| Edited :|
| cj wrote:
| They can easily make it not a scandal by changing things so
| applicants contribute to an open source project that they
| don't directly benefit from as a business. Easy solution
| treprinum wrote:
| Let's just make qdrant to pinecone/weaviate/redis/etc. data
| exporters, that would make the company super happy! Free
| labor benefiting their competitors.
| guappa wrote:
| Why pay when you can get free labour?
| infecto wrote:
| This sits somewhere in the middle for me. On one hand I
| probably would not do this exercise for most companies but I
| probably would for a company I was excited about. It kind of
| makes sense if they are getting high volume of applications,
| you definitely will miss great candidates doing this but does
| it not also servce as a self-selection for the type of
| candidate they want?
|
| Curious if the co-founder who was posting in here will share
| his take.
| lysecret wrote:
| Yes I agree. I think you could easily add if the pr is merged
| we reward it with some money etc and it would be perfect.
| prakhar897 wrote:
| To give more context from an employee's point of view [just
| my opinion and might be completely wrong]
|
| For a job switch, I need to spend time in three different
| stages:
|
| ---------------
|
| Preparation:
|
| Leetcode (Blind 75) : 150 hours
|
| System Design + DBMS + OS + Networking : 100 hours
|
| Behavioural Questions (preparing STAR format answers): 10-20
| hours
|
| ------------
|
| Application:
|
| Avg time for sending 500 applications: 20 hours (Assuming 1
| application every 2.5 minutes)
|
| -------------
|
| Interviews:
|
| Let's say I got 25 callbacks and 10 of them asked for
| takehome.
|
| Person to person interviews time: 25 * 3 = 75 hours
|
| Takehomes: 10 * 6 hours = 60 hours.
|
| -------------
|
| All in all, I'm already spending 415 hours of unpaid work to
| get x% of salary increment. Not including the side projects
| or hackathons we may need.
|
| So having a takehome exercise asking to make an active
| contribution to the company is....bad. Sure i can reject it
| but not everyone will. which is what led us into the multiple
| rounds of algo interviews hellhole.
|
| I apologize if what I'm saying is harsh. All I want is for
| leadership to see us as humans with families and not monkeys
| jumping through hoops.
| infecto wrote:
| Not harsh and I don't disagree. On the flip side though if
| they have a large volume of applicants and they are a sub
| 50 company right now, it probably does not matter what kind
| of hoops they make people jump through, they will most
| likely identify candidate that match their fit. What I am
| saying is that nobody is wrong in this situation.
| codetrotter wrote:
| If it's a job that wisely does not emphasize leetcode, you
| can skip those 150 hours of that.
| ativzzz wrote:
| yea but a lot of the super high paying ones do, so it's a
| small price to pay for a large paycheck
| azinman2 wrote:
| I'm in a privileged position and have never done
| leetcode.
| jonathankoren wrote:
| I have worked at multiple FAANGs and even small startups.
| I have never once seen anyone care about leetcode, GitHub
| punchcards, stack overflow score, or any of the social
| media stuff people boast about here. Literally none of
| this fits in to any evaluation rubric.
|
| All it says is your hobby is programming.
| randomdata wrote:
| But you must remember that they don't want you. They
| already have more applicants than they can handle.
|
| The trouble is that the leadership does see you as human,
| which results in them trying to say "Go away! You are not
| welcome here." as politely as possible.
| srackey wrote:
| 415 hours??
|
| Well this is the problem with tech hiring. Not convinced an
| IQ test or other "unstudyable" exam wouldn't be better.
|
| Hell maybe nepotism was the way to go all along. Do other
| industries require 10 weeks of full time practice to get a
| gig?
| 7thaccount wrote:
| My industry is related to the power grid. I've worked at
| two amazing companies. The first had me come out for a 3
| hour interview for a summer internship where they
| accessed my work ethic and culture fit. Once I graduated,
| I was immediately given an offer letter. The second job
| required most of a day to interview and I had to prepare
| a PPT and then got an offer. I also interviewed for
| another gig that did like four one-hour interviews spread
| out across a month. What software developers do sounds
| like absolute hell. My industry has very high demand and
| very low supply of experienced candidates at the moment
| though.
| randomdata wrote:
| It is not a problem with tech hiring in general, just
| hiring where there are millions of people lined up down
| the street vying for the same position. To be sure, the
| job will still most likely go to a friend or relative,
| but if you are willing to jump through insane hoops you
| _might_ also be considered. But it is to be taken as a
| hint that says: _" Unless you are extra super sure that
| you are so special that we can't turn you down, don't
| waste your time, or ours."_
|
| Most other jobs, including Mom & Pop Tech Co., are happy
| if anyone applies at all and will take what they can get.
| esafak wrote:
| They want candidates who care about _their_ product, not
| people who merely rank companies by compensation, subject a
| constraint on time spent preparing. If you don 't
| particularly care what you are working on, you would be
| better off at a big company.
| pclmulqdq wrote:
| On homework problems for jobs, I have a strict policy of "no
| more than 4 hours of free work, and I retain full copyrights
| to that work." A lot of people are picking up on the first
| clause of that policy, and companies seem to be adapting, but
| the second clause still isn't common yet.
| epistasis wrote:
| When the company is asking for open source contributions to
| an open source code base, as it is in this particular case,
| that second clause is clearly a deal breaker.
| shizcakes wrote:
| I don't see a single such submission in their pull requests,
| open or closed, in the past couple weeks.
| simonw wrote:
| Read that comment more closely: they didn't ask for PRs to
| one of their own repos, they asked for a brand new repo to be
| created, and a link to that repo to be emailed to them.
| nijave wrote:
| You could always release and license under BUSL or AGPL or some
| other business non friendly and relicense if you got the job.
| LudwigNagasena wrote:
| You can simply publish it without any license, in which case
| you reserve all rights.
| sgc wrote:
| Nobody is going to hire you if you can't follow basic
| instructions.
| osigurdson wrote:
| This is actually kind of brilliant.
| anonylizard wrote:
| Most of the prestigious and elite indie game firms (Those that
| pay very well, fully remote, have a hugely successful product
| that can be sold for decades) basically only hire modders into
| their team.
|
| Like, you had to have actively developed mods for them, for
| free, for years, and be famous in the community, then they'll
| hire you (If you want).
|
| This works because the working conditions there are far far
| better than your average game company. And probably much more
| fun than say a bank.
| mpawelski wrote:
| What concrete companies you are talking about?
| sireat wrote:
| For some reason this does not seem as exploitative as most of
| the interview circuit elsewhere.
|
| Over the years I've read/heard plenty of stories (here on HN
| and elsewhere) of people getting hired for their open source
| contributions to some stack that some company is
| using/developing.
|
| So here I am willing to give some slack here to Qdrant. They
| get extremely qualified candidates who can jump right in, and
| candidates get told the rules of the game up front. It feels
| fine?
|
| Surely much better than fake take home tests, whiteboard tests,
| leetcode onslaught, and 7 layers of interviews.
|
| So if creating a high quality repo is 60-90% of your job
| interview that seems pretty good. As long as they are not
| ghosting high quality contributions that is.
|
| I will change my view if they get 20 high quality connectors
| out of this and noone gets hired from that pool of candidates.
| anonzzzies wrote:
| Offtopic: Is there a good OSS mixed (vectors + traditional) that
| can be embedded in our own solution and allows storing indexes in
| a pluggable kv storage? Besides rolling one, I cannot really find
| anything. Rust or Go would be best.
| anonzzzies wrote:
| The sourcecode is very readable of this product. And good
| license, no agpl or worse stuff.
| treprinum wrote:
| Why is AGPL bad? It would prevent Amazon from taking it from
| the founders and making money off it without giving anything
| back, like they did to dozens other products.
| JoachimS wrote:
| It also makes it less interesting for other potential
| customers to use it. Reducing the potential market is
| probably not what a VC funding a startup wants.
| menaerus wrote:
| IMHO it looks arcane to understand or to debug, and with most
| likely a lot of negative performance implications, due to its
| shared-ptr-in-disguise all-over-the-place design.
| $ git grep "Arc<" | wc -l 451
|
| It could be probably related to the fact that the main author
| of the codebase is coming from the Java/Scala world. Or perhaps
| it's the Rust safety guarantees.
| nemothekid wrote:
| Qdrant is a async Rust project, so there will be lots of Arc.
| Rust safety guarantees doesn't really let share references
| across threads haphazardly.
| menaerus wrote:
| Something being async doesn't imply shared-ptr design. But
| perhaps this is what Rust makes you to to achieve its
| safety guarantees?
| nemothekid wrote:
| Yes; in many cases, if there are two tasks that hold some
| piece of data, the compiler cannot, at compile time, when
| to free that data, as it doesn't know which of the two
| tasks will finish first.
|
| That means the lifetime of that object must be tracked at
| runtime, with a garbage collector. This is where you get
| Arc (or shared-ptr).
|
| I'm not sure how you would safely solve this in C++, but
| FWIW, scylladb is a high performance database that also
| makes use of shared_ptr.
| rvz wrote:
| Well deserved funding round for a company that underpins most of
| the AI hype happening all over the place and probably always
| overlooked by many analysts.
|
| Let's see what they can do in a year or more with that new
| capital.
| infecto wrote:
| I am excited to see how the vector search space plays out. Most
| of my work is not constrained by a low latency chat type user
| experience and I have not touched most of the vector search apis.
| I wonder what the difference is between competitors. The way I
| picture it is everyone is starting up their own Elasticsearch
| hosted solution and while there are some differences in
| functionality, the real bet is cost and scale.
| ankit219 wrote:
| I think alpha lies in how good the embedding space is rather
| than which db you use to store and retrieve. A typical tradeoff
| between accuracy and performance, and here accuracy will be
| more important in many cases esp for businesses and
| enterprises. With that, and existing database providers
| introducing their own support for vectors, this space might be
| commoditized in near term.
|
| Re embeddings, you would likely get better results if you train
| your own embeddings model. A popular approach is ColBERT, which
| anecdotally outperforms vector search in border cases[1].
| Second is training an embedding model using initial layers of
| an LLM. [2]. In Colbert's case once it's trained, you dont need
| a db to store the vectors.
|
| [1]: https://twitter.com/arjunkmrm/status/1744741903646773674
| [2]: https://huggingface.co/intfloat/e5-mistral-7b-instruct
| infecto wrote:
| I agree with you. I was ignoring the accuracy/performance
| tradeoff. Even in that space while there is certainly a lot
| of innovation left, there is already so much that is
| available commercially open source. If that holds true, you
| are really left with competing on price and scale in the long
| run.
| lettergram wrote:
| Not to knock Qdrsnt, but generally the whole "vector search
| database" rush is insane.
|
| I've been working with vectors for over a decade; particularly
| with embeddings used in AI. We're talking projects from 100k to
| 100B+ records, used for AI applications
|
| Postgres, particularly with pgvector and derivatives, can handle
| to millions of records very rapidly no problem. It's very cheap,
| scales great, and is accurate.
|
| I'm sure some of these open source solutions are improvements.
| That said, weigh vendor lock in, cost, risk and in the end it
| usually makes very little sense.
| francoismassot wrote:
| Congrats to Qdrant's team, $28M for a Series is really nice.
|
| There are a lot of OSS vector search databases out there, we
| could probably list the main ones:
|
| - Qdrant: https://github.com/qdrant/qdrant
|
| - Weaviate: https://github.com/weaviate/weaviate
|
| - Milvus: https://github.com/milvus-io/milvus
|
| What else?
| andre-z wrote:
| These are the major ones, correct.
| mritchie712 wrote:
| We use pgvector which if you're already using postgres should
| be in the running for your use case.
|
| I also like https://github.com/lancedb/lancedb
| tajd wrote:
| This website for comparing vector database solutions might be
| handy https://vdbs.superlinked.com/
| lmeyerov wrote:
| It's funny taking a numbers view. The most popularly used might
| not even be these, but vector indexes in existing popular OSS
| DBs and storage systems people are already using. Afaict
| earliest would be faiss on disk and vectors in opensearch &
| elasticsearch, and I'd be curious how say databricks, pgvector
| and other big ones are getting picked up now that they are out.
| Most of these supported fast & large-scale indexes even early
| on (ivfpq, ...) by wrapping faiss and friends. ~All OSS DBs we
| use now, esp managed, have or are getting vector indexes.
|
| Another one most similar to qdrant we track internally is
| lancedb. They are clever by supporting an embedded
| architecture, so an architectural reason to prefer over most
| existing OSS DBs. In our survey 2 years ago, we predicted
| specialized vector DBs having regular OSS DBs be the elephant
| in the room, and missed embedded as a fundamentally different
| category: https://gradientflow.com/the-vector-database-index/ .
|
| (Good luck to qdrant! I'm happy they waited before raising,
| hopefully this means they can operate more healthily than
| otherwise and easier to maintain the discipline to do that!)
| jillesvangurp wrote:
| There are a few more. Pinecone comes to mind.
|
| And then there are traditional databases and search products
| that are integrating vector search capabilities as well:
| Postgres, Elasticsearch, Opensearch, Solr.
|
| They each have their limitations of course but the 28M round
| suggests a moat that I'm not seeing that clearly in terms of
| tech. What's so special about qdrant relative to their
| competition?
|
| At least they are Apache licensed for now. So, that's nice. But
| that also means e.g. Apache Lucene could borrow some code from
| them to beef up their vector search capabilities. Which would
| benefit Elasticsearch, Opensearch, and Solr which all depend on
| Lucene.
|
| Which raises the question what the point is of QDrant long term
| and why investors are betting on this as opposed to other
| things.
|
| It seems to me that the main challenge with vector search is
| inference cost (at index and query time), not storing the
| vectors. A secondary concern is the vector comparisons at query
| time. A good way to cut down on that is to reduce the overall
| result set using traditional search or query mechanisms. In
| other words, you need
| lsaferite wrote:
| Is Pinecone OSS?
|
| I ask because this was the statement from the PP
| There are a lot of OSS vector search databases out there, we
| could probably list the main ones ... What
| else?
| utopcell wrote:
| Pinecone is not OSS.
| manishsharan wrote:
| I think there will be enough of market to justify a few more
| dedicated VectorDB vendors.
|
| From the enterprise perspective, which of these vendors
| proved the best combination of security, availability,
| performance and pricing will matter. when we run benchmarks
| on our (self hosted) LLMs, we do not a clear idea of where we
| have bottlenecks and we end up assuming its the GPU/memory.
| And our pilot implementation will never go into production as
| the security model is nearly non existent in our
| implementations; the execs AND qa are getting the same RAG
| outputs. It is all very new to us and our teams. If a vendor
| can outperform its competition in our tests and show credible
| security model with segmentation of knowledge, that would be
| the choice.
| internet101010 wrote:
| I couldn't agree more. I would add reproducibility to the
| list of important things, above everything else you
| mentioned. I looked into Vanna after seeing it on here
| because generating SQL code by only embedding the schema
| and business logic seems like a nice, quick middle ground
| that doesn't require embedding an entire database. However,
| 88% accuracy in generating the correct query isn't good
| enough for deployment at the organization-level. "Give me
| sales for the last quarter as of end of prior month" should
| return the same result for everyone, without exception.
| epistasis wrote:
| It's fascinating to see the diversity of vector databases! I've
| chosen to prototype with two, ChromaDB, and LanceDB, based on
| the ease of using embeddings with them, and had not even heard
| of these others here. I'm also very excited to go through
| VectorHub's table of databases:
|
| https://vdbs.superlinked.com
|
| (discovered from sibling comment here:
| https://news.ycombinator.com/item?id=39103322)
| daveed wrote:
| I think Activeloop(YC) is too:
| https://github.com/activeloopai/deeplake/
| rgbrgb wrote:
| I've been looking at this one to embed in desktop apps
| https://github.com/unum-cloud/usearch
| alfalfasprout wrote:
| Sure, but frankly it's historically _very_ hard to build a
| business around a specialized database, especially if you have
| competitors that are even 80% as good but free.
|
| The cases where I've seen this work are when the DB offers
| something way ahead of what their competitors offer. For
| example, KDB+ was historically unrivaled when it came to ultra
| high performance time series storage and Aerospike is very hard
| to beat for extremely high performance multi-node K/V.
|
| Otherwise there's little to stop a larger company from offering
| the OSS competitor to your DB as a service for a lower cost and
| invest eng resources to close the gap.
| shenli3514 wrote:
| Chroma looks good. https://github.com/chroma-core/chroma 10k+
| stars, very easy to use, and can be used as an embedding
| database
| morgango wrote:
| Elasticsearch: https://www.elastic.co/platform
| tw1984 wrote:
| I don't think such business model is going to last. There is no
| reason for AI giants like OpenAI to stick with such external
| "vector databases". There is not much technical stuff there.
| Unless you want to argue that "vector searching" is just some
| labor work when compared to AI, in that case, sure.
| manishsharan wrote:
| There are huge segments e.g. banking, insurance,legal, which
| are wary of using OpenAI and they would much rather host their
| own LLMs. I think these vector databases will find a ready
| market in this segment
| tw1984 wrote:
| Tell me what makes you believe that those big techs are not
| going suit those "banking, insurance,legal" orgs by providing
| them their own LLMs? For example, ever heard about github
| enterprise? you pay a stupid amount, github setup everything
| almost identical to the public github, just on your servers
| for your employees. Why big techs won't do the same here?
|
| Those high profit margin part of the LLM business is for big
| players _only_ , they don't burn hundreds of billions to
| offer you opportunties to cut their profit by capitalizing on
| their _core_ business.
|
| Communism doesn't exist in high tech. People don't work their
| xxx off to pave ways for your free lunch for life.
| manishsharan wrote:
| I wanted to respond to you but your hostile tone implies
| you are not looking for a conversation.
| mritchie712 wrote:
| OpenAI uses Qdrant for chatgpt and a few other products.
|
| https://news.ycombinator.com/context?id=38611608
| guappa wrote:
| And has enough money to fork it at any time.
| tw1984 wrote:
| or just build from scratch with assistant from AI - then
| market it as built by AI for AI.
| esafak wrote:
| They could just ask it to do the same to their own
| product and generate GPT-5
| nemothekid wrote:
| Is the idea that big players will get tired of paying a license
| to a company like Qdrant and write their own database? I just
| don't really see why they would do that - if Qdrant is similar
| in complexity to any standard DB, it's like asking why doesn't
| Apple just rip out MySQL and write their own Apple DB.
|
| I can see them replacing it if Qdrant isn't able to scale to
| their needs - thats why we ended up with Dynamo, Spanner,
| MyRocks. However its likely that its probably easier to just
| acquire the team - like Apple did with Foundation - and become
| project stewards than trying to invent a new datastore to save
| pennies.
| spullara wrote:
| qdrant is trivial compared to a normal database.
| CamperBob2 wrote:
| _There is not much technical stuff there._
|
| I know that, and you know that, but tell it to a jury of East
| Texas hayseeds.
| tw1984 wrote:
| We have to be honest - "vector" database is a _low_ tech stuff
| when compared to today 's AI. You shouldn't be expecting to walk
| into the battle of AI, which is arguable the most important one
| in our life time, to dig a chunk of significant profit from major
| AI players' pocket by just having some low tech stuff. They use
| external "vector databases" _for now_ because they don 't want to
| invest R&D resources on such non-key issues _for now_.
|
| _for now_ is the keyword here.
|
| When the company grow to 10k or 30k people, there will be teams
| competing for visibility, someone is going to build their inhouse
| "vector database" to get his/her slice of the pie. Do you still
| believe that any AI major player is going to reply on some
| external vector databases?
| coffeebeqn wrote:
| Are in-house databases that common? I thought generally we've
| found as an industry that to be a great thing to purchase. I do
| wonder how many will need anything other than the vector
| support in their already existing Postgres instances though
| tw1984 wrote:
| > I do wonder how many will need anything other than the
| vector support in their already existing Postgres instances
| though
|
| exactly! if there is a real & strong demand, we'd be seeing
| open source ones get upgraded & ready in months. it is more
| like just one of those "I want to build something easy in the
| core but fancy in its name to get some quick VC $"
| Prosammer wrote:
| My understanding is that these vector search databases are
| generally used by people who want to use an existing AI and
| extend it with RAG etc. Was anyone ever expecting major AI
| players to use a tool like this as you are suggesting?
| tw1984 wrote:
| > use an existing AI and extend it with RAG etc
|
| companies like openai will fill the gap by offering such
| features out of the box. there is no logical reason why and
| how a big AI tech giant is going to take all hard work and
| letting someone else to take the profit by ignore the last
| mile issue. in fact, openai has already released such APIs in
| their last devday event.
| PheonixPharts wrote:
| Can you name me a single 10k-30k people company that has their
| own internally built relational database? Their own internally
| developed document database? I've never seen this in my career.
|
| I don't even know any 10-30k people companies that build their
| own _search_ , most I've known use elastic search or lucene.
|
| It's had to imagine that a few of these vectordb companies
| don't establish themselves as the standard solution, being the
| equivalent to MongoDB in their space. The other competent
| players will very likely get acquired.
|
| Certainly these vectordb companies are in a better long term
| standing that the bajillion companies rushing to build products
| that are just calling an API endpoint at the end of the day.
| yujian wrote:
| Good on them, I know the crustaceans are out here happy about
| this raise for a Rust based Vector DB!
|
| (now I'm gonna plug what I work on)
|
| If you're interested in a more scalable vector database written
| in Go, check out Milvus (https://github.com/milvus-io/milvus)
| andre-z wrote:
| The open-source benchmarks show different results. Feel free to
| make a PR to improve. ;) https://qdrant.tech/benchmarks/
| softwaredoug wrote:
| Someone has to ask the question: How many vector DBs do we really
| need? How do the vector DB companies differentiate themselves?
| And why do we need a company at all when there are increasingly
| awesome open source options?
|
| I genuinely ask - there are a lot of other problems in the RAG,
| fine tuning, AI/LLM, retireval space, to solve. And more and more
| vector retrieval is, while not 100% solved, at least is something
| the community has a grasp on the tradeoffs. Solved to the point
| that squeezing a bit more recall out of vector retrieval isn't
| the problem anymore.
| inertiatic wrote:
| >Solved to the point that squeezing a bit more recall out of
| vector retrieval isn't the problem anymore.
|
| I think this is a bit of a strawman. I don't think recall is
| the main point these systems are trying to sell us on, it's
| more about robustness and ease of use compared to building
| something inhouse or using a lower level library to build a
| system on top of it just for this small part of your overall
| project/product (be it RAG, search, whatever).
|
| I guess Lucene-based solutions, while very mature overall in
| terms of engineering, lagged behind this functionality (out of
| caution, trying to build what's going to be long term useful)
| and are also perceived a bit too cumbersome. So these stores do
| make sense, I think. The core functionality is nothing too
| complex (at least HNSW), but hiding it behind a stable black
| box with just a few inputs and levers, has value for people
| that are likely to use these stores.
| sanp wrote:
| Agree but then the same argument applies to RDBMSs and multiple
| vendors seem to be doing OK in that space. I think it
| ultimately comes down to "stuff" (sales journey, price, support
| etc.) other than the technology itself. I am sure any RDBMS can
| meet most of the requirements of any given customers (in most
| cases) but we still see customers buying across vendors.
| esafak wrote:
| qdrant _is_ open source. Being open source is not in opposition
| to running a company; it is part of their strategy.
|
| There is still work to be done in vector databases. None of the
| products have perfected hybrid search yet, for example, and
| performance varies a lot between products; they are not
| fungible.
| hartator wrote:
| > For example, it can automatically map 'frontend engineer' to
| 'web developer'
|
| Small revolution indeed.
|
| Ref: https://qdrant.tech/use-cases/
| avereveard wrote:
| I really don't understand that sample the similarity capability
| is provided by the external embeddings model not by quadrant
| per se, unless they have some proprietary embeddings.
| minimaxir wrote:
| Correct, but that one-pager is more aimed toward project
| managers than engineers. Marketing copy is weird like that.
| ancorevard wrote:
| Honest question, how long before EU makes it unattainable for
| Qdrant to remain in Germany/EU?
| wahnfrieden wrote:
| What's the best vector db for text similarity that can run in
| browser front ends too?
| spullara wrote:
| Honestly there is no reason, except huge scale, to have a
| separate vector db. Every normal database and search engine now
| support vector search.
| braza wrote:
| Outside AI and LLMs, there are some solid use cases for those
| Vector Search Databases? Maybe I am not seeing something, but
| it's hard to see it gaining traction outside tech companies.
| esafak wrote:
| Vector databases enable semantic and similarity search. What
| company does not need that?
| beernet wrote:
| Companies that don't want/can build it by themselves, so the
| majority of enterprises. It's a nice series A by the numbers,
| at the same time, generating relevant revenue will very
| likely not happen (given the valuation at this round was
| probably around 200MEUR). It's hype all over but can't blame
| them, would do the same I guess.
| yding wrote:
| Congrats! Amazing milestone.
___________________________________________________________________
(page generated 2024-01-23 23:01 UTC)