🚀 Autonomous AI Recruitment Manager — Backend (FastAPI + LangGraph + Gemini)

Autonomous backend system that semantically evaluates resumes, identifies top candidates, proposes interviews, sends them for HR approval, and—only after approval—automatically schedules Google Calendar meetings and emails candidates.

This backend powers the full product.
Frontend repo: https://github.com/parv18050212/ai-recruiter-app

---

📌 TL;DR — Run in 30 seconds

git clone https://github.com/parv18050212/hr_agent
cd hr_agent

python3 -m venv .venv
source .venv/bin/activate  # Windows: .\.venv\Scripts\activate

pip install -r requirements.txt
cp .env.example .env       # Add DATABASE_URL, keys, etc.

# Add Google OAuth credentials.json in root
python get_token.py        # Creates token.json after browser auth

uvicorn hr_agent.app.main:app --reload

Open → http://127.0.0.1:8000/docs

---

🎯 What This Backend Does

1. Resume Understanding

• Upload PDF
• Text extracted + embedded using Google Gemini
• Stored inside Postgres pgvector
• Compared against job vectors → similarity score

2. AI-Driven Candidate Ranking

• LangGraph pipeline interprets candidate-job match
• If strong match → create PENDING interview record

3. Human‑in‑the‑Loop Safety

Agent cannot book interviews automatically.
HR must approve via /pending-interviews/{id}/approve.

4. Autonomous Interview Booking

After approval:

• Google Calendar booking
• Meet link creation
• Automated Gmail email to candidate

5. Candidate Portal

/my-applications/{email}

6. Feedback Loop

HR feedback stored for future model improvements.

---

🧠 System Architecture

              ┌───────────────────────────────┐
              │           React UI             │
              │ (candidate + HR dashboards)    │
              └──────────────┬────────────────┘
                             │
                             ▼
                   ┌───────────────────┐
                   │     FastAPI       │
                   │ REST + Handlers   │
                   └───────┬───────────┘
                           │
     ┌─────────────────────┼─────────────────────┐
     ▼                     ▼                     ▼
[Resume Parser]   [Gemini Embeddings]   [LangGraph Agent]
 Extract text      Store vectors in     Decide fit → propose
 in Python         Postgres(pgvector)   interview

                   ┌───────────────────┐
                   │  HR Approval UI   │
                   └─────────┬─────────┘
                             ▼
                   [Google Calendar API]
                   [Gmail Automated Emails]

---

🛠️ Tech Stack

Core

• FastAPI (REST API)
• LangChain + LangGraph (agent workflow)
• Gemini Embeddings
• PostgreSQL + pgvector
• Google Calendar API
• Gmail API

Utilities

• Python 3.10+
• Uvicorn
• Pydantic
• SQLAlchemy

---

⚙️ Setup / Environment Variables

.env.example includes:

DATABASE_URL="postgresql://user:pass@host/db"
GOOGLE_API_KEY="AIza..."
LANGCHAIN_TRACING_V2=true
LANGCHAIN_ENDPOINT="https://api.smith.langchain.com"
LANGCHAIN_API_KEY="lsv2_..."

GOOGLE_OAUTH_CREDENTIALS="credentials.json"

Enable pgvector extension

CREATE EXTENSION IF NOT EXISTS vector;

---

🔑 Google OAuth Setup

Steps

1. Google Cloud Console → Create Project
2. Enable:
   • Google Calendar API
   • Gmail API
3. OAuth → External → Add test users
4. Create Desktop OAuth Client → download credentials.json
5. Place file in project root
6. Run:

python get_token.py

Generates token.json.

---

📡 API Examples

Create Job

curl -X POST "http://127.0.0.1:8000/jobs" -H "Content-Type: application/json" -d '{"title":"Backend Engineer","description":"Python + FastAPI","location":"Remote"}'

Upload Resume

curl -X POST "http://127.0.0.1:8000/jobs/1/candidates" -F "name=Parv" -F "[email protected]" -F "resume=@/path/resume.pdf"

Get Pending Interviews

curl http://127.0.0.1:8000/pending-interviews

Approve Interview

curl -X POST http://127.0.0.1:8000/pending-interviews/42/approve

Candidate View

curl http://127.0.0.1:8000/my-applications/[email protected]

---

🧩 Design Notes

• Agent never books without HR approval
• Google token.json required only once
• Vector dims depend on embedding model
• Minimal PII stored
• Embeddings cached where possible

---

📅 Roadmap

• ATS integrations (Greenhouse / Lever)
• Better UI for HR review
• Full audit logging
• Explainability layer for ranking
• Multi-round interview planning

---

📄 License

MIT License.
Feel free to fork, extend, or deploy.