production deployments · live international clients · systems that don't break
| 🎓 | Pre-Final Year @ HBTU Kanpur — B.Tech, Class of 2027 |
| 💼 | Freelance ML Engineer — live international client (AgTech, Colombia/Venezuela) |
| 🔧 | Building — LLM/RAG systems · Computer Vision · Production MLOps |
| 🧠 | Obsessed with — Multi-agent pipelines · RAG evaluation · Systems that scale |
| 🎯 | Belief — Ship it. Instrument it. Fix it. Repeat. |
| 🌎 | AgroIA Ganadera (Colombia / Venezuela — AgTech) |
| YOLOv8s aerial livestock detection · mAP@50 = 0.731 · 2,714-image dataset | |
| Live HuggingFace Spaces + FastAPI REST deployment · CPU inference @ ~1s/frame on 2K imagery |
| Project | What it actually does | Numbers | |
|---|---|---|---|
| 🧠 | Socra — AI Startup Interrogator | 5-agent LangGraph pipeline · Startup Tribunal mode · Clerk auth · Razorpay payments · 9 security fixes pre-launch | $0.007/session · 52s · 3-provider fallback |
| 📚 | RAG Teaching Assistant | 241 lectures + 1,230 textbook pages · HyDE · cross-encoder reranking · multi-turn memory | Faithfulness 0.773 · ~200ms latency |
| 🔬 | Skin Lesion Classifier | EfficientNetB2 · Grad-CAM explainability · 10k+ dermoscopy images · 7 classes | 87%+ val accuracy |
| 💳 | Credit Risk Predictor | 307k applicants · 57M+ rows · 157 engineered features · Optuna 100 trials · MLflow tracking | ROC-AUC 0.782 |
| ⭐ | traceloop/openllmetry (7k★) — PR #4226 under review |
| Added DeepSeek provider instrumentation · captures chain-of-thought as span attributes · 121 tests passing |
open to remote ML Engineer internships