debnil = {
"name" : "Debnil Dey",
"role" : "AI Engineer π€",
"location" : "π Kolkata, West Bengal, India",
"focus" : [
"Large Language Models",
"Reinforcement Learning",
"Computer Vision",
"Clinical AI",
],
"learning" : ["RAG", "Multimodal LLMs", "AI Agents"],
"fun_fact" : "Built EM AR gloves in a browser π§²"
}- π Passionate about building AI that solves real-world problems
- π RL system achieving 36% cost savings in smart manufacturing supply chains
- 𧬠Built a CDSS with 91% recall for clinical liver disease prediction
- β‘ Created browser-native AR hand tracking with zero installs
- π¬ Ask me about LLMs Β· ML pipelines Β· Agentic AI Β· Computer Vision
π€ AI / Machine Learning
π₯οΈ Backend & APIs
π Frontend & Visualization
βοΈ Tools & DevOps
Multi-Agent Supply Chain Optimization with Reinforcement Learning β production-grade RL system for Industry 4.0
| Metric | Result |
|---|---|
| π° Cost Reduction vs Baseline | 36% |
| π‘οΈ Resilience Score (under disruption) | 0.998 / 1.000 |
| π¦ Order Fill Rate | 99.7% |
| π Training Scale | 100 episodes Γ 182K steps |
| π Dataset | 913K rows (multi-store demand) |
Browser-native AR Hand Tracking with real-time Electromagnetic Physics Simulation β no install required
- π€² Dual-hand real-time tracking β 21 landmarks Γ 2 hands via MediaPipe Vision AI (GPU-accelerated WASM)
- π§² Gesture-controlled EM force model β FIST (100% pull Β· cyan glow) Β· PINCH (35% precision Β· amber) Β· OPEN (gravity release)
- β‘ Inter-hand lightning arcs Β· PBR metallic objects Β· WebGL 2.0 rendering Β· 200-particle systems
- π Geometry mode β Heron's formula, Triangle Inequality Theorem, live circle projections with real-time spatial math
- π Zero install β runs entirely in browser, deployed via GitHub Actions CI/CD to GitHub Pages
Clinical Decision Support System for Liver Disease Risk Assessment β interpretable AI with SHAP explainability
| Model | Recall | Precision | AUC |
|---|---|---|---|
| β XGBoost (Primary) | 91% | 84% | 0.92 |
| Random Forest | 87% | 81% | 0.89 |
| Logistic Regression | 87% | 82% | 0.90 |
| Clinical Target | β₯85% | β₯70% | β₯0.80 |