Proof of Synergy

An AI communication gym. Practice real conversations by voice, get live coaching, and build a skill graph that persists across sessions. Includes a resume-based technical interview mode.

Stack

Next.js 14 (App Router), React 18, TypeScript, Tailwind. OpenTelemetry is used for optional end-to-end tracing. Gemini is the live conversation partner; Sarvam handles voice (speech-to-text, text-to-speech) and coaching summaries. Cognee is an optional semantic layer for the skill graph. Everything degrades gracefully: with no keys the app runs on a local conversation partner, heuristic coaching, and the built-in skill-graph engine.

Run

npm install
npm run dev        # http://localhost:3000

Other scripts: npm run build, npm start, npm test, npm run typecheck.

Config

Copy .env.local.example to .env.local and add your keys. Everything is optional; the app degrades gracefully without keys.

GEMINI_API_KEY=      # live conversation partner (dialogue, follow-ups, pushback)
SARVAM_API_KEY=      # voice: STT, TTS, resume OCR, and coaching summaries
COGNEE_API_URL=      # optional skill-graph semantic layer
COGNEE_API_KEY=
DEMO_MODE=           # true swaps in labelled sample data when a service is down; never in prod

See .env.local.example for optional model/voice overrides and defaults.

Features

• Voice practice with live transcription and auto-stop on pause.
• Technical interview: upload a resume (PDF, Word, or scan via OCR) plus an optional job description; questions are tailored to it. Hands-free (reads each question, opens the mic) and auto-ends after about 15 to 20 minutes.
• Skill graph: technologies actually discussed become skill nodes, grouped by category. Skills are credited only from what the candidate said.
• Session summary with metrics and coaching.

Observability (optional)

End-to-end traces of the interview flow can be sent to Arize Phoenix via OpenTelemetry — free, and off by default. Run Phoenix locally (docker run -p 6006:6006 -p 4317:4317 arizephoenix/phoenix:latest), set PHOENIX_COLLECTOR_ENDPOINT=http://localhost:6006 in .env.local, and each request shows up as LLM/chain/tool spans. See docs/observability.md for local vs. Phoenix Cloud setup and the full span map.

Layout

• app/ pages and API routes
• components/ UI
• lib/ core logic (skill graph, prompts, Sarvam/Gemini clients, resume parsing)
• server/ WebSocket signaling server for voice sessions