# Abdullah Al Mahmud

> CS major at Texas Tech University. Building AI systems, from satellite data visualization to robots aboard the ISS.

Based in Lubbock, TX, USA. This file is a machine-readable summary of
Abdullah Al Mahmud's background, kept in sync with https://abdullahalmahmud.me.
A structured version of the same data is available at https://abdullahalmahmud.me/resume.json,
and the full resume PDF at https://abdullahalmahmud.me/Abdullah_Al_Mahmud_Resume.pdf.

## Education

- Texas Tech University — B.S. Computer Science, Minor in Mathematics
- GPA 3.95 · Jan 2025 – May 2028

## Experience

### Software Engineer Intern — Dorsia
New York, NY · June 2026 – Present

- Building a live "drop" reservations feature designed to handle 30,000+ concurrent users by architecting a WebSocket pusher service on AWS Lambda that fans out availability the instant limited inventory opens.
- Gave members a single surface to browse and claim limited-availability reservations by shipping the in-app drops experience end to end on a Laravel GraphQL backend and a React Native client.

Technologies: TypeScript, React Native, Laravel, GraphQL, AWS Lambda, WebSocket

### Software Engineer Intern — Yobo AI
Ontario, Canada · June 2024 – Dec 2024

- Achieved 99% accurate tool calls, cut audio false positives by 25%, and raised user retention by 68% by architecting a low-latency voice interface with Twilio, OpenAI Realtime APIs, and PyTorch VAD.
- Cut manual data entry by 85% and lifted demo-to-trial conversion by 38% by engineering an automated ETL pipeline with a fine-tuned Vertex AI LLM to extract structured menu data from images and UberEats.
- Scaled backend to 10,000+ daily requests and cut API latency by 150ms by migrating from Firebase Firestore to PostgreSQL and architecting a cloud-native microservices stack on AWS (Lambda, API Gateway, RDS, S3).

Technologies: Python, OpenAI, Twilio, PyTorch, Vertex AI, PostgreSQL, AWS Lambda, API Gateway

### Software Engineer Intern — Megamind AI Inc.
Ontario, Canada · Jan 2024 – May 2024

- Trained a predictive neural network classifier with TensorFlow to analyze facial features achieving F1 ~92%.
- Implemented optimized edge-computing solution using YOLOv3 and Jetson Nano, reducing inference time by 60% for real-time detection.
- Developed an efficient wake-word detection ONNX model achieving F1 score of ~99.48% using Openwakeword.

Technologies: Python, TensorFlow, YOLO, Jetson Nano, OpenCV, ONNX

## Leadership

### External Technical Advisor — EWB Cornell
Ithaca, NY (Remote) · Aug 2026 – Present

- Mentoring the Engineers Without Borders Cornell software team on backend engineering and deployment, advising on how they structure services and get them shipped to production.

Technologies: Backend, DevOps, Deployment, Mentorship
More: https://ewbcornell.org/teams/software/

### Member of Technical Staff — CodePath
Lubbock, Texas · Jan 2026 – Present

- Trained TTU students to build and deploy production-ready Agentic AI applications by designing curriculum and leading hands-on sessions on generative AI fundamentals and design principles.

Technologies: AI, Agentic AI, Workshops, Leadership

### Section Leader — Stanford CodeInPlace
Stanford, CA · Apr 2025 – Jul 2025

- Instructed 23 students over 6 weeks in control flow, data structures, and algorithmic thinking after being selected from thousands of applicants to teach Stanford's CS106A Python curriculum.

Technologies: Python, Teaching, Data Structures, Algorithms

## Projects

### zeg
An on-device AI interviewer that runs technical screening calls on one Dell GB10 box and returns a 1-10 rubric score with quoted transcript evidence, so candidate audio never leaves the machine. The interview engine is a state machine over a full-duplex speech model with barge-in, a consent gate and a time budget, running two processes on an 80 ms frame clock over a loopback WebSocket. 1,250+ Python tests and eval suites cover scorer agreement, bias pairs, red-teaming and consent edge cases.
Technologies: Python, PyTorch, Hugging Face Transformers, pytest, WebSocket, WebRTC, Next.js, NVIDIA GB10
Live: https://zeg-app.vercel.app · Source: https://github.com/zero-abd/zeg

### teamree
A macOS Electron app that runs several coding agents in parallel, each in its own git worktree with real PTY terminals (node-pty, xterm.js), ranking every pane by failed, working, waiting or finished. The same runtime is exposed as a CLI over an owner-only local socket that returns one JSON document per command, so an agent can drive it while the window updates live. A self-hosted Cloudflare Worker relay lets agents on one machine securely access context from another, end-to-end encrypted with the Noise IK protocol so the relay only ever sees ciphertext. 3 releases shipped.
Technologies: TypeScript, Electron, React, xterm.js, node-pty, Cloudflare Workers, Vitest
Live: https://teamree.us · Source: https://github.com/zero-abd/teamree

### Appwing
A remote MCP server on Cloudflare Workers with 33 tools that let Claude fill internship applications, tailor resumes and track every application, behind Google OAuth with PKCE. A deterministic Chrome MV3 extension carries per-ATS selector libraries for Greenhouse, Lever, Ashby and Workday, so fills need no LLM and nothing submits until the user approves a screenshot. Also ships a tailored-resume pipeline with an anti-fabrication validator and a TeX Live WASM renderer.
Technologies: TypeScript, Cloudflare Workers, Durable Objects, D1, MCP, OAuth 2.1 (PKCE), React, Chrome MV3
Live: https://appwing.us

### Astrobee Controller
Placed 5th out of 400+ teams competing aboard the ISS by programming Astrobee Robot 6-DOF movement using Quaternions with the KIBO API to navigate waypoints and avoid keep-out zones. Achieved 97% accuracy detecting 8 tools with a TensorFlow model optimized to TFLite for Astrobee's Snapdragon 820, and used ZXing to decode AR codes for autonomous navigation.
Technologies: Java, Gradle, TensorFlow, LiteRT, ZXing, Docker
Source: https://github.com/zero-abd/Astrobee-controller

### biponi
A bilingual (English and Bangla) online store builder where AI drafts product listings from photos, on Next.js 15 and Supabase Postgres with row-level security across 54 migrations. Six payment methods (bKash, Nagad, SSLCommerz, aamarPay, manual mobile banking, cash on delivery) and five courier APIs sit behind provider registries, with webhook-driven order tracking. An LLM sales agent quotes prices and places orders against a deterministic server-side price floor, so the model can never discount below the merchant's limit. 1,500+ Vitest tests in CI.
Technologies: TypeScript, Next.js 15, React 19, Supabase, PostgreSQL, Turborepo, Vitest, Gemini API
Live: https://biponi.app

### Graphclaw
Won 1st place out of 200+ teams at JacHacks 2026 by building a graph-native multi-agent runtime in Jac where memory lives as a property graph and agents act as walkers routing to sub-agents that handle multiple workflows. Shipped support across 5 channels (Telegram, Discord, Slack, Email, WhatsApp) with a persistent graph memory model featuring session/turn tracking, confidence decay, and runtime skill + MCP server installation.
Technologies: Python, Jac, Jaseci, LiteLLM, Playwright, MCP
Live: https://graphclaw.vercel.app · Source: https://github.com/zero-abd/graphclaw

### draw2agent
Reached 5,000+ users with an MCP server that overlays Excalidraw on local dev pages, letting developers annotate UI directly and submit visual context (screenshot + DOM + annotations) to AI agents. Published 4 tools to the MCP Registry, enabling real-time sketch-to-code editing in Cursor, Antigravity, and Claude Desktop, with iPad support for remote drawing and live site updates.
Technologies: TypeScript, MCP, AI, Excalidraw, Node.js
Live: https://draw2agent.vercel.app/ · Source: https://github.com/zero-abd/draw2agent

### intendr
Built at the Ramp Hackathon: an MCP server that gives any AI agent a spend-capped wallet for the Orthogonal data API catalog and commerce, with guardrails that block over-cap spend and hold real-world purchases for user approval. Every charge is a single atomic Supabase Postgres RPC that debits the wallet and writes the ledger row, so concurrent agent calls cannot overspend. Published to npm as npx intendr, with a browser device-code sign-in.
Technologies: TypeScript, Cloudflare Workers, Hono, MCP, Supabase, PostgreSQL, React, npm
Live: https://intendr.vercel.app · Source: https://github.com/zero-abd/intendr

### Ortha
An AI chat app that finds and calls endpoints from Orthogonal's API catalog at runtime through four meta-tools (search, inspect, run, expand), with web search and scraping as fallbacks. Each tool result is distilled into a summary while the full payload stays in a per-conversation Durable Object with SQLite, pulled back only on demand. Per-workspace spend caps run as an atomic conditional UPDATE in D1, with a per-provider circuit breaker and 370+ Vitest unit tests.
Technologies: TypeScript, Cloudflare Workers, Durable Objects, SQLite, D1, React, Vitest, Gemini API
Live: https://ortha-web.pages.dev · Source: https://github.com/zero-abd/ortha

### BuffAdvisor
Won 1st place in the AMD AI PC track at HackCU 11 with a CU Boulder campus assistant that runs entirely on a local AMD AI PC. Quantized DeepSeek-R1-Distill-Llama-8B with AMD Quark and served it through ONNX Runtime GenAI, with FAISS retrieval over campus documents and answers streamed over Server-Sent Events.
Technologies: Python, Flask, ONNX Runtime GenAI, AMD Quark, LangChain, FAISS, React
Live: https://buffadvisor.vercel.app · Source: https://github.com/zero-abd/BuffAdvisor

### High-Speed Visual Deterrent
A closed-loop pan-tilt tracker built for a research project under Dr. Changwon at Texas Tech: frames stream from a Jetson Nano to a laptop, MediaPipe Face Mesh locates targets, HSV filtering finds laser dots, and PD control drives four servos. Moving all vision to a FastAPI laptop service left the Jetson a camera and servo I/O server, with a web panel for live tuning, manual calibration and debug overlays.
Technologies: Python, FastAPI, OpenCV, GStreamer, MediaPipe, NumPy, Jetson Nano

### StudySync
AI-powered syllabus PDF analysis system using RAG to extract course schedules, assignments, and grades. Built unified dashboard with real-time chat assistant, grade calculator, and task management. 1st Place at Tidal Hack 2025.
Technologies: Next.js, TypeScript, Flask, LangChain, LLMs
Live: https://studysync-abd.vercel.app · Source: https://github.com/zero-abd/StudySync

### Roam-E
WRO autonomous robot car with computer vision running on NVIDIA Jetson Nano. Features PID-based wall following, traffic signal detection, and real-time navigation. Bronze Medalist at World Robot Olympiad 2023.
Technologies: C++, Python, Jetson Nano, OpenCV, PID Control
Source: https://github.com/zero-abd/Roam-E

### Artemis+ Lunar Habitat Simulator
Built the web app and backend for Artemis+, a lunar habitat planner at NASA Space Apps Challenge 2025, embedding the team's Unity WebGL simulation in a React and TypeScript site. Added a Gemini mission assistant whose prompt carries all four mission designs, and serve the Brotli-compressed WebGL build with a stateless assistant route on Vercel, where each visitor brings their own Gemini key.
Technologies: React, TypeScript, Vite, Node.js, Express, Google Gemini API, Unity WebGL
Live: https://artemis-plus.vercel.app · Source: https://github.com/zero-abd/artemis-plus-lunar-habitat-simulator

### StorePal
AI shopping assistant combining voice interaction with ElevenLabs ConvAI, interactive store mapping, and Pinecone vector search for intelligent product location. Built for HackUTA 2025.
Technologies: React, FastAPI, ElevenLabs, Pinecone, WebSocket
Live: https://storepal-app.vercel.app · Source: https://github.com/zero-abd/StorePal

### llmmcp
A powerful Model Context Protocol (MCP) server that provides AI agents with real-time access to the latest LLM documentation (Gemini, Claude, OpenAI), enabling accurate API usage without hallucinations.
Technologies: TypeScript, MCP, AI, LLMs
Live: https://llmmcp.vercel.app/ · Source: https://github.com/zero-abd/llmmcp

### DocLess
Built the React and TypeScript frontend for DocLess at HackRice 2025: Auth0 sign-in, dashboard, patient list, profile and settings screens. DocLess turns a nurse-patient conversation into a structured visit note, keeping every patient record in the clinician's own browser and running the AI step on the visitor's own key or a local Ollama model.
Technologies: React, TypeScript, Vite, Tailwind CSS, Auth0
Live: https://docless-app.vercel.app · Source: https://github.com/zero-abd/DocLess

### Landsat-Lens
Real-time NASA Landsat satellite tracking and data visualization platform. Tracks Landsat 8 and 9 live from Celestrak orbital data on interactive Leaflet maps, and pulls real surface reflectance, NDVI and surface temperature for any clicked point from the Landsat Collection 2 archive.
Technologies: React, TypeScript, Leaflet, satellite.js, STAC API, Vercel
Live: https://landsat-lens.vercel.app · Source: https://github.com/zero-abd/Landsat-Lens

### AuthLens
Blockchain-powered video authentication platform ensuring integrity of CCTV evidence. Stores cryptographic hashes on Ethereum for tamper-proof verification of video authenticity, timestamps, and ownership.
Technologies: Python, Solidity, FastAPI, Ethereum, React
Live: https://authlens-app.vercel.app · Source: https://github.com/zero-abd/AuthLens

## Skills

- Languages: C, C++, Java, Python, TypeScript, JavaScript, Rust, Jac, SQL
- Frameworks: React, Next.js, Angular, Vue, Flask, FastAPI, Django, Node.js, Ruby on Rails
- AI/ML: TensorFlow, PyTorch, OpenCV, YOLO, OpenAI, Twilio
- Agent workflows: Claude Code, Cursor, Orca, Hermes Agent, OpenClaw, MCP, Agent Loops, CI/CD Review
- Tools: Linux, Git, Docker, Redis, AWS, Google Cloud, Vercel, DigitalOcean
- Databases: PostgreSQL, Firebase, MongoDB, Qdrant, Pinecone
- Hardware: Jetson Nano, Arduino, Raspberry Pi

## Honors and Awards

- 1st Place — JacHacks 2026 (UMich) (2026)
- Champion, AMD AI — HackCU 11 (2025)
- Champion — Tidal Hack 2025 (2025)
- 5th Crew Award — 4th KRPC by JAXA (2023)
- Bronze Medalist — World Robot Olympiad (2023)

## Links

- Website: https://abdullahalmahmud.me
- GitHub: https://github.com/zero-abd
- LinkedIn: https://www.linkedin.com/in/abd-zero/
- Email: connectwithalmahmud@gmail.com
- Resume (PDF): https://abdullahalmahmud.me/Abdullah_Al_Mahmud_Resume.pdf
- Resume (JSON): https://abdullahalmahmud.me/resume.json