AlperMind: I built a complete AI SaaS and it didn't sell
AI sales assistant that answers using your company documents. Multi-tenant, SSO, Stripe, per-agent RAG. Try it here and read why it failed.
What AlperMind was
AlperMind was an AI sales assistant. The idea: sales teams waste hours digging through price lists, datasheets and terms buried in PDFs and emails. AlperMind reads those documents and answers instantly, citing where each fact comes from.
Above you have a demo with a fictional company. Ask anything from the list and notice that every answer points to its source document. That was the product.
What I built
Not a prototype. A complete SaaS, ready to charge:
- Streaming AI chat over each company's documents (RAG with per-agent vector stores).
- Multi-tenant with Owner, Admin and Member roles.
- Specialised agents per area: sales, technical support, after-sales, each with its own documents.
- Per-agent permissions: you decide which rep sees what.
- Enterprise SAML SSO, email login and OAuth.
- Stripe payments and subscriptions, including plan sync.
- Bilingual ES/EN, Sentry observability, Jest and Playwright tests.
Stack: Next.js 15, React, TypeScript, PostgreSQL with Prisma, OpenAI, NextAuth and Tailwind.
Why it failed
This is the part worth reading.
1. The market caught up while I was building
When I started, "chat with your company documents" was a differentiator. Within months ChatGPT Team, Microsoft Copilot and Google Gemini shipped it built-in, inside tools companies already paid for. Competing against a free feature of your email provider is not a fight a solo developer wins.
2. I built the platform before having a customer
SAML SSO, multi-tenancy, billing: big companies ask for that. But no big company was waiting for me. I spent months on enterprise infrastructure without validating that anyone would pay.
3. Selling B2B software is a different job
The code was the easy part. The hard part is the sales cycle: meetings, pilots, IT asking where data is stored, procurement asking for three quotes. That is a full-time job, and I was coding.
4. Trust in data beats technology
A small business won't upload its price list to a tool from someone it doesn't know, however good the demo. Big platforms win on trust, not quality.
What I learned (and do differently now)
- Customer first, code second. I don't build anything without someone who will use it from day one.
- The problem, not the technology. Nobody cares if you use RAG. They care about answering their customer faster.
- What applies lives on. Everything I learned about AI over documents I now use in real company projects, where the customer exists and the problem is clear.
Got a problem like that: documents nobody finds, answers that take days? Let's talk.
Frequently asked questions
What is an AI sales assistant?
A chat that answers sales questions (prices, lead times, specs) using the company's internal documents instead of generic internet knowledge.
What is RAG?
RAG (Retrieval-Augmented Generation) means first retrieving the relevant passages from your documents and handing them to the AI model, so it answers with your data and cites the source.
Can something like this be built for my company today?
Yes, and far cheaper than in 2025. For most SMBs a simple setup on top of the tools they already use is enough, no custom SaaS needed.