Gautham

AI voice trainer, helps real estate agents perfect their listing presentations and get better at talking to clients.

Scored 214,000 global energy facilities on how likely a drone attack is to succeed and how bad it would be if it did, fusing six public incident, infrastructure, and population datasets into a weighted two-part risk model. Shipped an interactive map and a natural-language chatbot for querying site-specific risk.

Peer-to-peer fishing gear rental app

Drone defense for critical infrastructure (Oil refineries, Data centers, Power plants, etc.), built around a metal mesh net system. For more info contact gautham.hari@aegisgrid.net

A dance learning app that uses pose estimation to compare your movement against a reference clip and tell you where you're off. Won $3,000 at Pitch Dingman.

An orchestrator watches model uncertainty, notices when a task looks like a new domain, and spins up a specialist from a meta-learned prior. Old specialists get frozen so they don't get overwritten, and inputs route by cosine similarity. Currently testing the idea on Split-CIFAR-100 before it goes near a robot.

Reference counting, mark-and-sweep, and stop-and-copy, implemented once in Rust and once in OCaml. Writing the same collector in two languages is the most I've ever learned from one assignment.

A Unix shell with piping and job control, plus a malloc implementation. Also an x86-64 battery meter simulator that reads memory-mapped device ports and unpacks bit fields by hand.

Feature detection and matching on drone spectral imagery using SIFT and SURF.

Active real estate license. Where the Listing Labs idea came from.