Senior Platform Engineer at PAM AI · Founder of Burki · Co-founder & former CTO of BiteBuddy
I build voice AI infrastructure and products people actually use: from conversation orchestration and telephony to the app, the backend, and the customer deployment.
Explore my work · Résumé · LinkedIn · Email
| Work | My ownership | Outcome / current stage |
|---|---|---|
| PAM AI | Senior Platform Engineer. In-house voice orchestration, conversation controls, provider routing, SIP, transfers, and failure recovery. | Systems handling 50,000+ calls/day; approximately $100k/month estimated vendor-cost savings. |
| BiteBuddy | Co-founder & CTO, Dec 2024–May 2026. Personally built the core voice, SMS, and web ordering product, backend, menu ingestion, and restaurant deployments. | 300,000 production calls, 10 restaurant deployments, $100k ARR. Recruited a five-engineer team. |
| Burki | Founder & engineer. A weekend replacement for a voice provider grew into a reusable phone-agent platform. | 5 agency customers in week one; a white-label UAE deployment handling approximately 20,000 calls/day. |
These are product and deployment outcomes; the public examples below are separate, self-contained demonstrations. Commercial and employer implementations remain private.
- Brimigo — pet discovery, mutual matching, real-time chat, and meetups. React Native / Expo, Supabase, PostgreSQL, and PostGIS, including approximate-location privacy and row-level access controls. Delivered through Apple App Store review.
- Kalbi — a private shared space for everyday moments, pings, countdowns, widgets, and small rituals. Currently in private beta.
-
GaariGar — first place in my sixth-semester Software Design and Analysis class. A Java/Spring Boot roadside-assistance platform with Android clients and web administration. Inspect layered architecture, notification-handler interfaces, a payment-gateway abstraction, and reusable query specifications. Design walkthrough · Live sample.
-
Voice Session Lab — synthetic conversation sessions for examining provider failover, interruption, tool idempotency, and transfer recovery. Executable scenarios, event traces, and tests.
-
TrajectoryShield — AI safety research on policy violations across agent tool sequences. Effect tracking, 300 authored fixtures, reproducible evaluation, and an honest audit of the evaluation limits. Explore the interactive trace demo.
-
Agent Trace Evals — evaluate an agent's tool actions against explicit policies, with evidence attached to each finding. Synthetic fixtures and reproducible reports.
-
Streaming + function calling in FastAPI — an API example for streamed model output and validated tool execution.
-
Hybrid search with PostgreSQL — vector retrieval, full-text search, and reciprocal-rank fusion.
-
RL Bandits — reproducible experiments with ε-greedy, UCB, and gradient bandits, including tests and comparison plots.
-
Burki documentation — public integration and architecture documentation for the voice platform.
I completed my M.S. in Computer Science at Texas Tech University in May 2026, working with Professor Akbar Namian on policy following and deceptive behavior in AI tool-calling systems. I focus on what an agent actually does, including the actions hidden behind a plausible final answer.
The TrajectoryShield research brief explains the approach, findings, and limitations. Code and reproducible examples are public; the paper has not been released.
Previously: engineering and team leadership at SHARE Mobility, and founding-engineer work at Doodhwala.
Core tools: Python, TypeScript, FastAPI, PostgreSQL, Redis, AWS, Docker, WebSockets, SIP, and React Native. My recurring interests are conversation lifecycle, failure recovery, evaluations, observability, and shipping the whole product.
Based in Northern Virginia / Washington, DC.