AI/ML Engineer. Architect. Tinkerer of brains—biological and artificial.

I specialize in building agentic AI systems that actually think, not just autocomplete. From CFG-driven LLMs to cloud-native MLOps pipelines, I design AI that’s composable, scalable, and a little dangerous (in a good way).

• Building multi-agent orchestration engines at FastpeerAI
• Designing LLM pipelines that obey rules, not vibes
• Co-developing Large Brain Models with Max Hospitals & IHBAS
• Making EEGs do things they were probably never meant to do

• ⚙️ HLGuidance → Grammar-constrained LLM generation
  → Built with GuidanceAI. Structured generation by leveraging Guidance's API to the fullest, a no-nonsense AI orchestration framework which gives just enough tooling to build an AI centric software without using 100s of over-engineered abstractions, coughs .. langchain .. coughs.

• ... When I find the time to properly arrange and publish here.

• Refactored LLM inference with CFGs and regex, doubling context recall and slashing latency by 90%
• Achieved 82% EEG classification accuracy for epilepsy detection in clinical settings
• Created a neural cellular automata library 10x faster in JAX—because why not

• 🎓 IIT Mandi | Double Major: CS & Bioengineering
  Neuroscience, deep learning, and other fun stuff.

• 📄 EEG-Depth-Image (ICPR Workshop '24)
  Depth map reconstruction from EEG—don’t ask how, just know it works (cosine sim: 0.81)

• 📄 Epilepsy Detection, Deconstructed (Under Review - IEEE Epilepsia)
  Tore apart the Bonn EEG dataset. Built something better.

• Languages: Python (can read my mind), Rust (just enough), C/C++ (if I must)
• AI/ML: PyTorch, LangChain, TensorFlow, Guidance, vLLM
• Infra: Kubernetes, Terraform, Argo, GCP, DigitalOcean
• Extras: FastAPI (pro), MNE, YOLOv8, JAX, Pyspark

• 🥇 Amazon ML Challenge – AIR 18
  Cracked product categorization with a vision-language pipeline and a 0.71 harmonic F1

• 🧠 INSERB Youth Conclave – 2nd Runner-up
  Invented a cognitive meta-language for aliens. The judges survived.

• 📫 [email protected]
• 💼 LinkedIn
• 💻 GitHub

Built with AI™ — just like your fridge manual and LinkedIn JDs, but sassier and twice as confused.*