Edge AI & on-device intelligence  ·  Firmware × ML  ·  Agentic systems & Observability

I build things from scratch to understand them — autograd from the chain rule up, vision operator by operator, ML that runs on real silicon. Write the naive version, prove it with tests, then reach for the library.

Optum (UnitedHealth Group) — Technology Development Program Intern

• Built an agentic pipeline that patches security vulnerabilities and CI failures unattended, then proves each fix against a baseline in a sandbox before it is allowed to merge — orchestrated on Azure Durable Functions, Service Bus, and Cosmos DB.
• Delivered end-to-end observability for a .NET platform — distributed tracing, KQL workbooks provisioned in Terraform, and a React dashboard for API health, errors, and dependency flow.

Ultrahuman — Firmware & AI/ML Developer

• Step-counting on raw accelerometer data (~95%) and a TFLite walk vs. non-walk classifier to kill false positives.
• Integrated into Zephyr RTOS firmware and shipped to production wearables over OTA.

• micrograd — scalar autograd engine built from scratch; the chain rule walked backwards over a DAG.
• NumPy-only Neural Net — nets in pure NumPy, hand-derived gradients, no framework; ends on ECG transfer learning.
• ImageLab (Julia) — classical computer vision from scratch, operator by operator. 1530 tests.
• EnergyMem — agent memory as convergence in a learned energy landscape.
• Sensor Drift & Data Poisoning — fog layer telling benign drift from adversarial injection.
• IMU Assistance-Dog Alert — collar that verifies a seizure-alert and escalates it.
• Fog-Based Vehicle Monitoring — edge–fog–cloud health; millisecond decisions stay local.
• Graph-Based Location Privacy — five privacy algorithms on GeoLife GPS traces.
• Semiconductor Yield & Defect — SECOM failure prediction, WM-811K defect-map CNN.
• AI Web Scraper — prompt-driven extraction on a local Llama 3.1.
• PG Suite — PGHub (4,500+ DSA problems), PG.Chess, PG.Play, blog.

First-principles essays on AI, intelligence, power, and work.

B.Tech CS (IoT) @ VIT Vellore · CGPA 9.93/10 · University Rank 1, all three years.

Understanding systems from first principles. Building to understand.

Carpe Noctem.