Srinjoy Roy
Srinjoy Roy builds production-grade ML systems that move from model to market.
Technical Stack
Selected Archival Works
MetaMark (SIH 2025 Winning Project)
AI-powered legal metrology compliance engine for e-commerce listing validation.
Nirvana Health Chain (HackHeritage 2024 Winning Project)
Anti-counterfeiting pharma supply chain system for drug safety and inventory trust.
Phishing Website Detection
End-to-end ML pipeline for phishing detection with automated training, evaluation, and deployment.
Medical OCR and NLP Intelligence
OCR pipeline for noisy medical documents with NLP extraction for medicines, dosage, and patient metadata.
AgriX Nirvana
Farm-to-fork agricultural supply chain transparency platform for authenticity, quality, and fair pricing.
Smart Green Campus
Real-time campus sustainability platform for energy, water, and waste optimization.
The Trajectory
Cybersecurity Intern
Designed and built a Transformer-based anomaly-detection WAF from scratch.
- ▹Designed and built a Transformer-based anomaly-detection WAF from scratch — a 5-stage pipeline (log ingestion, parser/normalizer, custom BPE tokenizer, 6-layer DistilBERT-scale encoder ~66M params, FastAPI + ONNX inference sidecar) that flags requests deviating from learned-benign traffic.
- ▹Generated a 500K+ entry labelled training corpus by deploying 9 intentionally vulnerable web apps across 7 tech stacks with Locust-based traffic simulation.
- ▹Architected non-blocking sidecar inference via an Nginx/Lua hook and async worker pool, plus a LoRA-based continuous retraining pipeline for zero-downtime model hot-swaps.
- ▹Built the platform's authentication system (login, signup, MFA) and core security/admin modules: IP blocking, geo-blocking, and maintenance mode.
Open Source Contributor
Diagnosed and fixed a Clipboard API bug causing UI crashes.
- ▹Diagnosed and fixed a Clipboard API bug causing UI crashes on Copy actions across MLflow's prompt, tracing, and evaluation modules in non-HTTPS deployments.
- ▹Designed a shared copyToClipboard() utility with Clipboard API and execCommand fallback, refactoring 5 frontend components with full unit-test coverage.
Team Leader and AI/ML Architect
Led MetaMark, an AI compliance system for legal metrology validation.
- ▹Architected OCR plus LLM compliance validation using Tesseract, Gemini, and LangChain.
- ▹Implemented regulatory rule validation with real-time trust score generation.
- ▹Owned model logic, system design, leadership, and final deployment.
- ▹Recognized with national television coverage for innovation and hackathon achievement.
Backend Developer and ML Engineer
Led ML pipeline design for anomaly detection in drug inventory supply chains.
- ▹Led ML pipeline design for anomaly detection in drug inventory supply chains.
- ▹Built backend services integrating ML inference and real-time monitoring.
- ▹Deployed production models with CI/CD, Docker, and cloud infrastructure.
Education
Beyond the Code
Why Should You Hire Me?
Don't just take my word for it. I build hardcore machine learning into interactive experiences. Choose your challenge.
01. The Grandmaster
I have trained a custom neural network on thousands of my personal Chess.com games. It thinks like me, it plays like me, and it is quite strong.
If it beats you, you hire me. Deal?
02. The Architect
Anyone can wrap a black-box API. I build entire neural architectures from scratch natively in the browser.
Step into the lab to train a network live and see real deep learning under the hood.
03. The Visionary
I can run cutting-edge computer vision natively in your browser using a custom-built WebAssembly Vision Transformer.
Write a word by hand and watch my Edge AI read it in real time.