class SanatWalia:
role = "AI/ML Engineer"
location = "Noida, India 🇮🇳"
education = "B.Tech + M.Tech (CSE) — JIIT Noida | Graduating Jun 2026"
focus = ["Agentic AI", "LLM Systems", "RAG Pipelines", "LLM Safety", "MLOps"]
currently = "Building multi-agent systems & MCP servers for production"
achievements = [
"🥇 National Winner — Smart India Hackathon 2024 (Govt. of India)",
"📄 IEEE Published — Face Morphing Defense @ PDGC 2024",
"🔬 KV-Probe Research — AUC 0.95, 94% jailbreak detection",
"🚀 Shipped to millions of users @ Bobble AI (97% latency reduction)"
]
def say_hi(self):
return "I don't build demos. I ship things that run in production. 🛠️"
|
Hierarchical multi-agent legal assistant with supervisor + specialized agents (research, chat, drafting). Reflection-based retrieval, persistent state, and full observability. Stack:
|
Novel method to detect harmful intent in LLMs via KV-attention probing — stops jailbreaks before output generation. Stack:
|
|
MCP server exposing 10 AI tools via Bearer token authenticated RESTful API. Real-time red flag detection, manipulation analysis, emergency resources. Stack: |
Full ML pipeline: ETL with Apache Beam → training → experiment tracking → CI/CD deployment. Fully reproducible and production-ready. Stack: |
| Achievement | Details | |
|---|---|---|
| 🥇 | Smart India Hackathon 2024 — National Winner | Built sub-7MB ONNX liveness + YOLO models for in-browser UIDAI edge deployment |
| 📄 | IEEE PDGC 2024 Publication | "Exploring StyleGANEX for Morphing Faces with Indian Ethnicity" |
HealthOrbit AI — Python Developer Intern Dec 2025 – Apr 2026
Multi-stage CPT medical coding pipeline (UAE HAAD/DHA). Built provider-agnostic LLM abstraction layer (OpenAI, Claude, Gemini).
Bobble AI — AI/ML Engineer Intern Jun 2024 – Dec 2024
Eliminated 97% sticker generation latency (5–10s → 0.3s). Shipped to millions of users. Cut server costs ~33%. Co-authored IEEE paper.
▸ LoRA fine-tuning + LangGraph agentic workflows
▸ KV-cache probing for LLM safety (active research)
▸ MCP server architecture patterns
▸ Sovereign AI & Indic language models
▸ Real-time streaming agentic pipelines
