Ask HN: No interest for my idea (private LLM and RAG for teams), how to proceed?

https://news.ycombinator.com/item?id=40960288

by namanyayg • 2 years ago

6 21 2 years ago

news.ycombinator.com

Hey HN,

I've been working on an AI + RAG B2B app for the past few weeks.

It's a AI chat interface for teams, to allow fast and private conversations with all company data (think Perplexity but searches on your company's data)

It uses open-source LLMs and has RAG integration with Jira, Slack, etc via LlamaIndex stored on the Weaviate vector DB.

Everything can be 1 click self-hosted for maximum privacy.

There are some similar products in this space so I know there's some demand.

But, I can't figure out a way to find potential users to talk to.

So far, I've tried Twitter ads and have found some people through my network, but in total it's only a couple of people who are actively talking to me.

What approach can work in this stage to get more people from my target audience to check out my product?

Or is this not a valid idea at all, and I should switch to working on something else?

(Link for the curious: prosgpt.com)

Cheers, N

newzisforsukas• 2 years ago

tailwind template site that looks like every other tailwind template site with four screenshots, not interesting. show me what it does in a more interactive way.

namanyayg10• 2 years ago

agreed (and upvoted). I built this within an hour to be able to collect contact information of interested people.

The main purpose of the app is allowing team members to search through the entire company knowledgebase

E.g. one can search through Jira to find all open tickets relating to a particular feature

Or, you can search through slack to find threads relating to a topic of your choice

Dutchie987• 2 years ago

Don't tell us, show us the value of your project!

Show how it would work in practice, show how it is better than normal Jira search. Show why I should use your project instead of the many, many others?

fallinditch• 2 years ago

This is a really exciting area to build a software business, there is huge potential.

Think about the audiences you need to get through to: potential customers yes, but maybe also systems integrators, consultants, agencies.

We can assume that there are many B2B service-oriented companies that have good client portfolios but do not have RAG-LLM-integration skills, some of these could be good targets to partner with.

You could consider focusing on a particular industry segment. Do an analysis to identify the most promising sectors. Build knowledge and contacts in your chosen sector(s).

More accurate target marketing in specialist media for that sector will give you more bang for your marketing bucks.

You could also contract with one or more independent sales reps who specialize in software sales. There are a bunch of platforms for 'commission based sales agents' that can link you up with software sales people.

I have more thoughts and will email you if you want to chat further.

namanyayg10• 2 years ago

Extremely insightful comment, thanks for the details. I would love to chat more.

Email: hi [at] nmn.gl

Discord: nmnyg

Cheers!

danenania2• 2 years ago

The description here in your HN post is clearer than the tagline on your website imo. My guess is it’s not clear to people exactly what your product does and why it’s valuable. You probably just need to iterate on your one-liner until it’s clear that people get it. “AI chat interface for teams”, “One-stop AI for teams”, and others like this are way too vague and generic I think.

Beyond that, emphasizing privacy while also advertising your use of models from OpenAI, Anthropic, Google, etc., is confusing. In my understanding, the central challenge of enterprise “chat with your company data” is how to do it without giving the data to cloud model providers. Self-hosting the server where the model calls are made doesn’t address this at all and I think you may risk engineers and security people viewing this as misleading marketing as a result.

namanyayg10• 2 years ago

Thanks for the comment. You're right about the tagline. I wasn't able to figure out something that's understandable and yet not overly technical. Can you elaborate on what would be a clear tagline for the app?

Regarding the privacy aspects: i should have made this clearer in the landing page, I also offer the option to use local/self-hosted llms through ollama. Thanks for the reminder, I'll update this soon.

danenania2• 2 years ago

Something like “Use AI with your company’s private data” maybe that makes it clearer what the app does. That’s not amazing but something along those lines perhaps. And be sure to also explain what makes it unique, since there are many other products in the space. What makes your product different from all the others? Simplicity? Integrations? Whatever it is, try to really hone in on that.

Makes sense about the local models. Definitely a good idea to make that clearer imo.

giantg2• 2 years ago

AI is a premium thing. The people in power to purchase stuff are going with big name providers because they don't really know what to look for in the details. This is especially true with the tight budgets tight now.

namanyayg10• 2 years ago

You're correct, but afaik there aren't many companies doing this featureset?

Plus, almost all of them will either be single model or involve heavy tracking which I can compete on

chatmasta2• 2 years ago

There are hundreds of companies pitching some variation of this product. That includes startups with a narrow focus and also large companies like Microsoft who are integrating CoPilot directly into their Office 365 offerings.

The issue is not the idea, which is about as obvious as “b2b spreadsheets” would have been in 1981. The value of the pitched product is self evident, if it works. But why should anyone choose to try it over the many relatively mature alternatives?

If you want to win any slice of the pie, you’re going to need to find some way to differentiate, and some kind of growth hack or flywheel to systematically expand. Preferably you can identify one vertical and use case where you can offer features that are sufficiently specialized for buyers to choose your product over generic solutions.

In other words: don’t try to be everything for everyone. Instead of wasting your time building pipelines for 1000 sources that nobody will use, pick a cluster of them and hitch your wagon to that. Go deep on that use case, and if you find success, only then begin to expand “wide,” deliberately and methodically.

I admire your ambition, but personally I’d consider attacking the problem from a different angle. This is a crowded space with well-funded competitors.

namanyayg10• 2 years ago

Thanks for the advice, what you've said makes sense and is a much needed reality check.

I'll try to figure out some ways to further niche it down.

If you have any advice on what angles pass this check I'd be happy to hear more about that.

chatmasta2• 2 years ago

Off the top of my head, maybe look for industries that are highly regulated and rely on specialized software that isn’t part of the same “Google Docs / Notion / Slack” combination supported by every RAG provider. Then your story can be that you’re an on-premises intelligence layer on top of these specialized data sources. Industries like law, healthcare, architecture, construction, finance, might all be suitable candidates.

I suggest you research those industries and make a list of their most used specialized software, which is unlikely to be connected already to a generic RAG solution. Then map out the industry size and participants: can you find some 10-50 employee companies to sell to? Locate their technology leaders (e.g. CTO or CIO) and ask them if you can do some user research for a product you think could help them (offer to compensate with a gift card). Prepare a list of questions, both specific and open-ended, and be prepared to have a discussion where you actively listen to them and probe deeper on their pain points. After the interview, ask to stay in touch and if they’d be interested in another session when you’ve developed a prototype they can play with.

Take this as hard earned advice from someone who spent five years paying for the mistake of “do everything for everyone:” DO USER RESEARCH. Everybody says it, hardly anyone does it. It’s tempting to think you know better, but trust me, you do not. It is the absolute best way to spend your time now, to figure out who you can sell to and what exactly to build for them, so you do not waste time building features and connectors that nobody will ever use. Also, if you do it right, those user research subjects could eventually become your first customers.

Ask HN: No interest for my idea (private LLM and RAG for teams), how to proceed?

Related Stories