Ramis 923ab1cf58 docs: finalize PodMan architecture and write full integration specs
Replaces v1 plan with locked architecture:
- Hermes orchestrator: POST /ingest → Gemini Vision → MongoDB → event detection → Gemini Live 2.5 voice via LiveKit Agents
- Four MongoDB collections: engineer_states, ownership_map, events, nudges
- Continual learning via ownership_map persisting across sessions

New files: docs/idea.md, docs/gemini.md, docs/livekit.md, docs/mongodb.md, docs/digitalocean.md, docs/demo-setup.md
Updated: README.md, docs/PLAN.md (12-hour build plan), database/README.md, infra/README.md, .env.example

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FFbfi4Cmb7BY75Wtne7bZn
2026-06-27 15:27:23 -07:00
2026-06-27 14:00:20 -07:00
2026-06-27 14:07:11 -07:00
2026-06-27 14:07:11 -07:00

PodMan — Real-time AI Team Coordination Agent

2026 AI Engineer World's Fair Hackathon — Track: Continual Learning

PodMan is an ambient AI teammate. Engineers join a pod room with earbuds in. Each engineer's browser PWA captures their screen every 30 seconds. PodMan watches, understands what each person is working on, detects coordination gaps — blockers, dependencies, duplicate work — and speaks up proactively before anyone has to ask.

"Carol, looks like you're waiting on auth. Alice is actively building it — hang tight." "Carol, Bob — Alice just got the auth endpoint running. You're clear to integrate."

Nobody sent a message. Nobody pinged on Slack. PodMan just knew.


How it works

  1. Engineers open the PWA in their browser and join a pod room
  2. PWA captures a screen frame every 30s via getDisplayMedia, POSTs it to Hermes
  3. Hermes (server-side orchestrator) calls Gemini Vision to extract structured context per engineer — current file, inferred task, terminal state
  4. Hermes writes context to MongoDB Atlas, updates the ownership map
  5. Hermes runs event detection across all engineers — dependency ready, blocker, duplicate work
  6. When an event fires, Hermes generates a spoken nudge and publishes it into the LiveKit room via Gemini Live 2.5
  7. Engineers hear PodMan through their earbuds

The ownership map persists across sessions — making PodMan faster and smarter each time the team works together.


Monorepo layout

Folder What
frontend/ React + Vite PWA — join pod, screen capture, nudge feed, live teammate status
backend/ Hermes orchestrator: /ingest endpoint, Gemini vision pipeline, event detector, LiveKit agent
database/ MongoDB Atlas schema — engineer states, ownership map, events, nudges
infra/ DigitalOcean App Platform deploy spec + Dockerfile
shared/ Shared TypeScript types
docs/ Full specs — read these first

Docs

File What
docs/idea.md Full concept, value prop, demo moment
docs/PLAN.md 12-hour build plan, team assignments, build order
docs/gemini.md Gemini Vision + event detection + Gemini Live 2.5 voice
docs/livekit.md LiveKit room structure, Hermes agent, voice delivery
docs/mongodb.md MongoDB collections, schemas, continual learning hook
docs/digitalocean.md DO deploy config, env vars, fallback plan
docs/demo-setup.md Demo laptop setup, pre-staging checklist

Prizes targeted

  • Best Gemini — Gemini Vision (screen understanding) + Gemini Live 2.5 (voice output via LiveKit Agents)
  • Best LiveKit — LiveKit is the real-time backbone for room presence and voice delivery
  • Best DigitalOcean — Hermes deployed on DigitalOcean App Platform

Quick start

cp .env.example .env
# fill in LIVEKIT_*, GEMINI_API_KEY, MONGODB_URI

pnpm install
pnpm --filter backend dev   # Hermes on :8787
pnpm --filter frontend dev  # PWA on :5173
S
Description
Ambient voice pair-programmer. 2nd overall of 95+ at the AI Engineer World's Fair 2026 hackathon.
Readme 2.2 MiB
Languages
TypeScript 84.7%
JavaScript 11.1%
Python 2.6%
CSS 0.8%
HTML 0.6%
Other 0.1%