Inspiration
As a true crime enthusiast who watches movies like Zodiac and shows like Mindhunter, I always wondered about going through old cold cases and doing something about them. But mostly, all you can do is consume the content passively, not work actively like a detective. That's why I built IAM Detective.
What It Does
IAM Detective is a gamified web app that gives you the feel of a real detective investigating a real case. You go through an evidence pinboard, unlock clues, and test your theories. The AI helps you interrogate suspects and witnesses, while an AI co-detective guides you throughout the case like a partner.
How We Built It
Built using Next.js on the frontend and FastAPI on the backend. For the AI features, I used DigitalOcean's AI stack - storing all the case files, police reports, evidence reports, Wikipedia articles, and books into a Knowledge Base, then connecting an AI Agent to it. The agent is prompted to behave differently depending on the API call context: as a suspect, co-detective, or witness. It's also called to generate the evidence board automatically.
Challenges We Ran Into
- Getting the AI Agent to function correctly across different personas
- Long load times for case files
- The evidence board not generating nodes properly
Accomplishments We're Proud Of
- The full case is solvable end-to-end
- User investigation progress is persisted across sessions
What We Learned
- How to leverage AI Agents for multi-purpose, context-aware interactions
- Connecting FastAPI with DigitalOcean's AI features
- Deploying on DigitalOcean App Platform
What's Next for IAM Detective
Scaling it to support any publicly available real-life cases.
Built With
- digitalocean
- fastapi
- nextjs
- openai
- postgresql
- python
Log in or sign up for Devpost to join the conversation.