Jahwanth Pulugujju

Sentra starts with a problem I want to understand properly: an AI agent can have valid credentials and still take an action it was never authorized to take. My unpublished SENTRA manuscript studies task-bounded runtime authorization in a controlled Python and SQLite payment model. Separately, the public Sentra Authority OS reference implements a vendor-approval workflow in TypeScript. The two implementations explore related questions; their evidence and results are separate. The decisions I care about are concrete: bind permission to the exact request, check it where execution happens, retain allowance when an outcome is uncertain, and keep enough evidence to examine a failure. The public reference uses deterministic policy, HMAC capabilities, gateway checks and hash-linked records. It still needs stronger production identity and tenant controls. I also build Intelli-Credit, a React and Supabase application for company and financial-document review workflows. Its next priorities are source-grounded extraction, authorization and integration tests. My DevTools documentation contribution to Symphony State was merged as upstream PR #60, and my VVLF work connects web development with founder coordination. My second unpublished manuscript is Calibration Scarcity Beyond Infinite Thresholds: An Audit of Counts, Ranks, and Label Resources. I am also studying Quantum Fundamentals and Advanced Algorithms through WISER and its program partners. I want to build systems whose decisions another engineer can inspect, reproduce and challenge.