Inspiration Founders spend months searching for problems worth solving. Meanwhile, millions of people post their frustrations on public discussion forums every day — a goldmine of validated pain points that nobody was systematically mining. We wanted to change that.
What it does PainScout automatically captures real user problems from public discussion forums, runs Amazon Nova as a true autonomous agent to research the market in real time across 5 live APIs (App Store, GitHub, HackerNews, Wikipedia, and forum search), and delivers a complete investor-grade analysis — TAM, competitors, funding intelligence, failure patterns, and an opportunity score — in under 60 seconds. One click then generates a working web app for the idea.
How we built it We built a forum monitoring app that captures posts across 27 categories. Posts flow into Supabase, where Nova Lite classifies them as real problems. Nova then acts as a genuine agent with native tool use support, autonomously deciding which APIs to call, how many times, and when to stop. The frontend is React + Tailwind, the backend is FastAPI, and the vibe coder uses Nova to stream a complete working HTML app in real time.
Challenges we ran into Getting Nova to behave as a true agent — not just reading pre-fetched data but autonomously driving the research loop — required careful tool design and message history management. Parsing Nova's output reliably across different response formats, and rendering generated apps live in an iframe without a build step, were the two biggest technical hurdles.
Accomplishments that we're proud of Nova autonomously calls up to 25 tool iterations, searches 5 different APIs with its own chosen queries, and synthesizes everything into a structured market analysis — all without any hardcoded logic telling it what to search. That's a genuine agentic workflow, not a scripted pipeline.
What we learned Nova's native tool use capability is fundamentally different from prompt completion — it enables real decision-making loops. We also learned that the quality of tool descriptions matters as much as the model itself; well-described tools produce dramatically better autonomous behavior.
What's next for PainScout AI Expanding beyond public discussion forums to ProductHunt, App Store reviews, and support tickets. Adding trend tracking so users can watch an opportunity grow over time. And making the vibe coder generate full deployable apps, not just prototypes — so the path from pain point to live product is completely automated.
Built With
- amazon-bedrock
- amazon-nova-lite
- amazon-web-services
- devvit
- fastapi
- github-search-api
- hackernews-algolia-api
- itunes-search-api
- postgresql
- python
- react
- render
- supabase
- tailwind-css
- typescript
- vite
- wikipedia-rest-api
Log in or sign up for Devpost to join the conversation.