Serhii Polishchuk

AI Founder — startup-validation platform

A multi-agent platform that validates startup ideas end to end — I built and marketed it solo, from idea to MVP.

Role
Solo — product, engineering & marketing
Stack
PythonFastAPICrewAILiteLLMQdrantRedisNext.jsReact
Architecture: web, Slack and Telegram channels feed a backend orchestrator (CrewAI, LiteLLM, Redis) over a Qdrant knowledge base and PostgreSQL, driving a research → analysis → synthesis agent cascade toward an MVP.

I ran this one solo — product, engineering, and the marketing to get it in front of people. It takes a raw startup idea and validates it end to end, from the first question to a working MVP, on a cascade-of-agents architecture.

The problem

Idea validation is mostly research: find the signal, structure it, and decide. Done by hand it is slow and inconsistent. The platform automates the research loop while keeping a human in control of the decision.

What I built

A FastAPI backend with a Qdrant knowledge base. Orchestration is a cascade of CrewAI agents — LiteLLM for provider routing and Redis Pub/Sub for coordination — so agents hand work down a chain of specialised roles. People reach it through a web app and through Slack and Telegram bots; a React frontend with auth and WebSocket live updates streams progress as it happens.

Decisions

Cascade over a monolith prompt. Specialised agents (research → analysis → synthesis) each do one job and pass structured output to the next. That made each step inspectable and cheap to improve in isolation.

Real integrations. Slack, Reddit, GitHub, GA4 and Stripe are wired in, so the platform runs against live sources and real billing from the start.

Marketing and launch

I owned the go-to-market too — positioning, the landing page and the launch channels — and put it in front of real users rather than a demo set.

Result

Testing the workflows with those users cut idea-research time by roughly 75%.