👋 Welcome to My Portfolio

This repository showcases my work, projects, and skills as an AI Research Engineer.
I’m passionate about building real-world AI systems that combine research and engineering to deliver impactful, deployable solutions.

---

📄 Resume

View My Resume

---

👨‍💻 About Me

Hello! I’m Shwetabh Biswas, currently working as an AI Research Engineer at AAV Innovation Labs & Johnson & Johnson.

Here are some key highlights from my work and journey:

• AI Engineer @ J&J & AAV Labs → Building AI systems like:
  • 📈 Innovation Graph: GraphRAG-powered collaborative ideation platform
  • 🏭 Resource Allocation: Forecasting model with SHAP & LIME interpretability for workforce optimization
  • 📑 Document Scoring System: Automated design document evaluator using Docling
• Researcher @ IISc Bangalore (2024–25)
  • Published at WACV 2025 (FRAUD-Net – 91.4% SOTA multimodal misinformation detection)
  • Built agentic AI pipelines with LangGraph (10× faster inference using KV caching)
  • Developed deepfake detection models & worked on Google Project Vaani for speech recognition

• Intern @ KissanAI → Fine-tuned Qwen-7B for agriculture-based VQA, improving performance by 2×, with public release on Hugging Face

• Early Projects → Built gallery agent, VAE image reconstruction pipelines, and person-detection systems integrating YOLO + FaceNet

I bring together academic rigor and industry-scale engineering to design AI that is practical, explainable, and impactful.

---

🎓 Education

• MSc Physics – IIT Jodhpur (2021–2023)
• BSc Physics – Center for Basic Sciences, Raipur (2018–2021)

---

🛠 Skills

• Languages: Python, SQL
• Frameworks: PyTorch, Hugging Face, LangChain, LangGraph
• ML/AI Domains: Computer Vision, NLP, Multimodal AI, Deepfake Detection
• LLM Techniques: RAG, PEFT, SFT, Agentic Workflows
• Deployment: FastAPI, Streamlit, Docker, CI/CD
• Tools: Git, SHAP, LIME, Docling

---

🚀 Projects

🔹 Person Detection and Cropping

YOLOv8 + FaceNet integration for face/person extraction, deployed with FastAPI & Docker.

🔹 Image Reconstruction Using VAE

CNN-based VAE for MNIST with LPIPS-based evaluation.

🔹 AI-powered Assistant

Multimodal assistant with Q&A, face recognition, object detection, and calendar/email integration.

🔹 Gallery Agent

Telegram bot for searchable photo galleries with OCR, captions, embeddings, and face recognition.

---

📚 Publications

• FRAUD-Net: Fraud News Detection using Uncertainty & Domain Adaptation (WACV 2025)
• Classifying Gravitational Waves with GMM and GLM techniques (PoS, 2023)

---

✍️ Blogs

• Understanding Multimodal Large Language Models
• Trying to learn what they “see” -PART : 1
• Trying to learn what they “see”: PART-2
• DeepFake Detection — Part 1 : Understanding Dual Attention Vision Transformers
• DeepFake detection — Part 2: Understanding LoRA based MoE adapter architecture

---

📫 Contact

• 📧 Email: shwetabhbiswas2305@gmail.com
• 💼 LinkedIn
• 🌐 Portfolio Website
• 📝 Medium
• 🐙 GitHub

---