Skip to content

README.md

Latest commit

 

History

History
141 lines (108 loc) · 4.81 KB

README.md

File metadata and controls

141 lines (108 loc) · 4.81 KB

LoopPrep — Multi-Agent Interview Coach

Paste a job description. Speak your answers out loud. Get a calibrated mock interview built around how that company actually interviews — with per-question content and delivery scoring, traced live in W&B Weave.

Built at AGI House x W&B Build Day, May 31 2026.

Tech stack

Layer Tool
Agent orchestration LangGraph
Backend API FastAPI + Uvicorn
Inference client openai SDK (W&B Serverless Inference endpoint)
LLM W&B Serverless Inference — see agent model breakdown below
Transcription Whisper local (openai-whisper, base model, CPU)
LLM tracing + eval W&B Weave
Session metrics W&B Core
Research Tavily
Frontend React + Vite (Node 20)
Containerization Docker + Docker Compose
Backend runtime Python 3.11

Agents

Agent Model Job
JD Parser meta-llama/Llama-3.1-8B-Instruct Extracts technical + behavioral signals from the JD
Research Agent meta-llama/Llama-3.3-70B-Instruct Looks up how the target company actually interviews
Format Agent meta-llama/Llama-3.3-70B-Instruct Builds the ordered question queue
Interviewer Agent meta-llama/Llama-3.3-70B-Instruct Runs the session, scores content per answer
Delivery Agent meta-llama/Llama-3.1-8B-Instruct Scores communication from Whisper transcript
Report Agent deepseek-ai/DeepSeek-V3-0324 Generates final session debrief

Project layout

backend/ app/ agents/ # LangGraph graph + node implementations routers/ # FastAPI routes services/ # LLM, Whisper, Tavily clients config.py # pydantic-settings main.py # FastAPI entrypoint observability.py state.py Dockerfile requirements.txt frontend/ src/ api/ # typed fetch client App.tsx Dockerfile docker-compose.yml .env.example

Quick start (Docker — Mac + Windows)

  1. Copy env template and fill in API keys: 
    cp .env.example .env
  2. Start backend + production frontend: docker compose up --build - Frontend: http://localhost:5173 - Backend API: http://localhost:8000 - API docs: http://localhost:8000/docs
  3. Dev mode with Vite hot reload: docker compose --profile dev up --build

W&B MCP server (Claude Code)

This repo ships a pre-configured W&B MCP server so coding agents (and you) can query Weave traces, W&B runs, and evals directly from Claude Code. It is committed and auto-approved — no per-developer approval prompt:

  • .mcp.json — defines the hosted wandb server, authenticated with ${WANDB_API_KEY} (no secret committed).
  • .claude/settings.json — auto-trusts it via enabledMcpjsonServers: ["wandb"].

Each developer only needs their own W&B API key in Claude Code's environment. The easiest way is a personal, gitignored .claude/settings.local.json:

{ "env": { "WANDB_API_KEY": "your-wandb-key" } }

(Alternatively, set WANDB_API_KEY as an OS environment variable.) Then open Claude Code in the repo root and run /mcp — wandb should show connected.

Local development (without Docker)

Backend

cd backend python -m venv .venv source .venv/bin/activate # Mac/Linux

.venv\Scripts\Activate.ps1 # Windows PowerShell

pip install -r requirements.txt cp ../.env.example .env uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Frontend

cd frontend npm install npm run dev Vite proxies /api → http://127.0.0.1:8000 (see vite.config.ts).

API overview

Method Path Description
GET /api/health Liveness check
GET /api/ready Readiness check
POST /api/sessions Create session (runs ingest pipeline)
GET /api/sessions/{id} Get session state
POST /api/sessions/{id}/transcribe Upload audio → Whisper transcript
POST /api/sessions/{id}/hint Trigger Socratic hint path
GET /api/sessions/{id}/report Get final report, close W&B run

Agent graph

START → jd_parser → [research ‖ format] → interviewer_loop → report → END

Interviewer loop: present question → record audio → transcribe → score content + delivery → advance queue.

Environment variables

See .env.example. All four required before the backend will start:

  • WANDB_API_KEY
  • WANDB_ENTITY
  • WANDB_PROJECT
  • TAVILY_API_KEY

Cross-platform notes

  • Paths: temp audio files use tempfile (not hard-coded /tmp).
  • Docker volumes: named volume for node_modules avoids Mac/Windows bind-mount perf issues in dev profile.
  • Line endings: repo uses LF; Git on Windows should use core.autocrlf=true (default).