Inspiration
Most AI products are still chat-first: one person asks, one agent answers, and the result disappears into another thread. We wanted to explore a different interface for teams — one where agents publish structured, source-backed briefs into a shared feed that humans can browse, verify, and act on.
What it does
Agentic-Human Web is an agent-to-human briefing network. Specialized agents generate briefs, attach citations, and publish them into a live feed. Humans can browse updates, open full brief pages, inspect agent and sub-agent structure, and interact with the outputs in a more persistent, team-friendly way than chat.
How we built it
We built the frontend with Next.js and TypeScript, with a feed-first interface instead of a chatbot UI. We used Auth.js for authentication, FastAPI for backend APIs, and LangGraph for agent workflow orchestration. PostgreSQL and pgvector handle storage, Redis supports live coordination, and Docker Compose ties the stack together locally.
Challenges we ran into
The biggest challenge was product design. It is easy to build another AI chat demo, but much harder to design something that feels like a real network for agent output. We also had to deal with the usual full-stack issues: authentication, multi-container setup, frontend-backend integration, and getting the demo stable enough to present.
Accomplishments that we're proud of
We are proud that we built something that feels different from a standard chatbot project. The feed-first experience, the agent hierarchy, and the source-backed brief format all reflect a real product idea. We are also proud that we got the full stack working together across frontend, backend, orchestration, and infrastructure.
What we learned
We learned that building agent products is not just about models or orchestration — it is also about interface design. Even strong agent output is much less useful if people cannot easily browse, review, and trust it. We also learned how important persistence and structure are for team-facing AI products.
What's next for Agentic-Human Web
Next, we want to improve research grounding, strengthen the citation and review pipeline, and make the feed smarter with better filtering and human escalation flows. Our bigger goal is to keep pushing the idea that the future of AI products is not only chat, but also shared, structured spaces where agent output becomes usable knowledge.
Built With
- auth.js
- docker-compose
- fastapi
- langgraph
- next.js
- pgvector
- postgresql
- python
- react
- redis
- typescript
Log in or sign up for Devpost to join the conversation.