PrepAI is an AI-powered interview preparation app that converts a job description + resume into a personalized mock interview with real-time scoring and a coach-style final report.

It is designed for a hackathon-style demo: text-first reliability, voice as an optional enhancement, and simple, explicit contracts.

• Paste a job description and upload/paste your resume
• Get a Revise Mode topic list (top gaps/must-haves)
• Take a mock interview (2 questions in the current prompt config)
• Answer via:
  • Voice (local Whisper STT)
  • Text (always available)
• See per-question evaluation:
  • Clarity / Correctness / Depth / Structure (0–5)
  • Strengths, missing points, and a recommendation
• Get a final report:
  • Overall grade and averages
  • Strengths / areas to improve / revision plan
  • Humanized by an LLM (same schema)

The system revolves around three core data contracts implemented in backend/schemas/:

• ExtractedProfile (backend/schemas/profile.py): job requirements + candidate profile + gaps
• InterviewScript (backend/schemas/script.py): questions + ideal answer outline + rubric
• EvaluationResult (backend/schemas/evaluation.py): per-question scoring + feedback

The aggregated output is:

• FinalReport (backend/schemas/report.py)

• UI: Gradio (multi-page flow with custom CSS) — ui/gradio_app.py
• Backend API: FastAPI — backend/app.py
• Orchestration: LangGraph — backend/graph.py
• LLM: OpenAI via LangChain — backend/tools/llm_client.py
• Voice (optional):
  • STT: local Whisper (openai-whisper) — backend/tools/stt_whisper.py
  • TTS: ElevenLabs — backend/tools/elevenlabs_tts.py
• Resume PDF extraction: pypdf — ui/components.py

flowchart LR
  UI[Gradio UI\nui/gradio_app.py] -->|"calls Python functions"| LG[LangGraph\nbackend/graph.py]
  API[FastAPI\nbackend/app.py] --> LG

  LG --> EX[Extractor Agent\nbackend/agents/extractor_agent.py]
  LG --> SB[Script Builder Agent\nbackend/agents/script_builder_agent.py]
  LG --> EV[Evaluator Agent\nbackend/agents/evaluator_agent.py]
  LG --> CO[Coach Agent\nbackend/agents/coach_agent.py]

  EX --> LLM[OpenAI via LangChain\nbackend/tools/llm_client.py]
  SB --> LLM
  EV --> LLM
  CO --> LLM

  UI -.optional.-> STT[Whisper STT\nbackend/tools/stt_whisper.py]
  UI -.optional.-> TTS[ElevenLabs TTS\nbackend/tools/elevenlabs_tts.py]

Notes:

• The UI currently imports and calls backend.graph directly (in-process).
• FastAPI endpoints exist for external clients and testing. The included run script starts both.

backend/
  app.py              # FastAPI endpoints
  graph.py            # LangGraph state graphs (step-by-step + end-to-end)
  agents/             # extractor / script builder / evaluator / coach
  schemas/            # contracts (Pydantic)
  tools/              # llm client, STT, TTS, similarity
ui/
  gradio_app.py       # Gradio UI (multi-page interview flow)
  components.py       # UI formatting + PDF text extraction
docs/
  PrepAI_flowchart.svg
  demo_script.md
scripts/
  run_local.sh

git clone https://github.com/PRONGS-CHIRAG/InterviewPrep.git
cd InterviewPrep

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

cp .env.example .env

Fill values in .env. Do not paste keys into chat or commit them to git. This repo already ignores .env via .gitignore.

• OPENAI_API_KEY
• OPENAI_MODEL (default: gpt-4o-mini)

• ELEVENLABS_API_KEY
• ELEVENLABS_VOICE_ID (defaults to a safe voice id if blank)

Whisper runs locally; if you use voice STT, you may need ffmpeg installed on your system.

./scripts/run_local.sh

Open the Gradio URL printed in your terminal (usually http://127.0.0.1:7860).

Backend:

uvicorn backend.app:app --reload --port 8000

UI:

python -m ui.gradio_app

Set in .env:

DEBUG_AGENT_OUTPUT=true

When enabled, these agents print their prompt and parsed JSON payload to the terminal:

• extract_profile (backend/agents/extractor_agent.py)
• build_script (backend/agents/script_builder_agent.py)
• evaluate_answer (backend/agents/evaluator_agent.py)

FastAPI provides stable endpoints in backend/app.py:

• GET /health
• POST /intake → IntakeRequest → ExtractedProfile
• POST /script → ExtractedProfile → InterviewScript
• POST /evaluate → {question, answer} → EvaluationResult
• POST /report → {evaluations} → FinalReport
• POST /run_full → {intake, answers} → {profile, script, evaluations, report}

• Flowchart diagram: docs/PrepAI_flowchart.svg
• 2-minute demo script: docs/demo_script.md
• Deep technical overview: overview.md
• Project brief/spec: Context.md

• Never commit .env or share API keys publicly.
• Resume and JD text are processed locally by the app; LLM calls send only the prompt content required for generation/evaluation.