AI-Powered-Job-Recommendation-System-using-LangChain-Gemini

AI-Powered Job Recommendation System (Streamlit + Gemini)

GitHub Pages landing: https://ayush245101.github.io/AI-Powered-Job-Recommendation-System-using-LangChain-Gemini/

Overview

A Streamlit web application that recommends personalized jobs or internships to students by combining user input (skills, experience, preferences) and resume parsing with a Retrieval-Augmented Generation (RAG) pipeline powered by a lightweight vector store and Google’s Gemini.

Key Features

• Upload resume (PDF/text) or manually enter skills & experience
• Skill extraction and normalization
• Lightweight job postings ingestion (sample dataset + pluggable live scrapers placeholder)
• Lightweight vector store using hashing-based embeddings (no FAISS needed)
• Semantic retrieval over job descriptions and skills
• Gemini LLM (google-generativeai) for personalized recommendation reasoning & concise explanations with safe fallback
• Application links and deadlines: shows “Apply” button when available, displays “Apply by” date, flags expired roles, and indicates days left (<= 7 days as “closing soon”).
• Streamlit UI with interactive filtering & export

Project Structure

app.py
src/
  ingestion/
    load_jobs.py
    resume_parser.py
  retrieval/
    vector_store.py
  pipeline/
    recommend.py
  utils/
    settings.py
    logging.py
data/
  sample_jobs.csv
notebooks/
  exploration.ipynb (placeholder)
docs/
  REPORT_TEMPLATE.md
.env.example
requirements.txt

Getting Started

1. Clone & Install

pip install -r requirements.txt

2. Configure Environment

Copy .env.example to .env and fill in your Gemini API key.

Environment variables:

• GEMINI_API_KEY (required for LLM ranking)
• EMBEDDING_MODEL (optional)
• VECTOR_STORE_PATH (optional)
• SERP_API_KEY (optional, reserved for future job-source integration)

Note: .env is git-ignored; never commit secrets.

3. Build Vector Store

The first app launch will automatically build a lightweight vector store from data/sample_jobs.csv if it doesn’t exist.

4. Run App

streamlit run app.py

Or via VS Code task: open Command Palette -> Run Task -> “Run Streamlit App”.

If GEMINI_API_KEY is not set the app will still work using a heuristic fallback (skill overlap + similarity scores) with an info banner.

Deployments

Option A: GitHub Container Registry (GHCR)

This repo is configured to build and push a Docker image to GHCR on pushes to main (and tags v*.*.*).

• Image name: ghcr.io/<owner>/ai-powered-job-recommendation-system-using-langchain-gemini:<tag> (e.g., ghcr.io/ayush245101/ai-powered-job-recommendation-system-using-langchain-gemini:latest)
• Workflow file: .github/workflows/docker-publish.yml

Run locally with Docker:

docker run --rm -p 8501:8501 -e GEMINI_API_KEY=$Env:GEMINI_API_KEY ghcr.io/ayush245101/ai-powered-job-recommendation-system-using-langchain-gemini:latest

Option B: Streamlit Community Cloud

1) Fork this repo 2) Create a new app in Streamlit Cloud pointing to app.py 3) Add a secret called GEMINI_API_KEY (if you want LLM ranking)

• In Streamlit Cloud → App → Settings → Secrets: add GEMINI_API_KEY="your_key"

Option C: Any VM / Render / Fly.io

• Build with Dockerfile provided or run via Python directly
• Expose port 8501; set GEMINI_API_KEY for LLM mode

Option D: Run directly on GitHub (Codespaces)

1) Open this repo in Codespaces (Use this template → Open in Codespaces) 2) The dev container installs Python dependencies automatically 3) Start the app in the Codespaces terminal:

  export GEMINI_API_KEY=your_key   # optional
  streamlit run app.py --server.port=8501 --server.address=0.0.0.0

4) Accept the forwarded port prompt to open the app in your browser

Notes on Streamlit “on GitHub”

• GitHub Actions runners are ephemeral; they aren’t suited to host a persistent Streamlit server.
• Use Codespaces for an interactive session or Streamlit Community Cloud for hosting.
• A Streamlit Smoke Test workflow is provided to verify the app boots on a runner.

Sample Input

• Skills: Python, Machine Learning, NLP
• Preferences: Remote, Internship, Location: “India”

Sample Output (Excerpt)

1. ML Research Intern @ Alpha Labs
Reason: Matches ML + NLP focus; internship; remote-friendly; aligns with your Python + transformer experience.
Confidence: High
 Meta: Remote • Internship • Apply by: 2025-11-15 (12 days left)

RAG Flow

1. Ingest job postings -> clean -> embed -> store locally (hashing-based)
2. Parse resume / user input -> extract skills & preferences
3. Form retrieval query & metadata filters
4. Retrieve top-K similar job vectors
5. Re-rank & summarize with Gemini using prompt template (or heuristic fallback)
6. Streamlit displays ranked recommendations

Extending (Next Steps)

• Add real job source scrapers (LinkedIn/Google Jobs) via official APIs or compliant scraping
• Add feedback loop (thumbs up/down) -> store in simple SQLite + fine-tune rerank logic
• Add caching layer (Redis) for repeated queries
• Implement multilingual support

Disclaimer

Live scraping may violate ToS; integrate only with proper permissions or official APIs. Provided code ships with a static sample dataset.

License

MIT