Transistors -> Logic -> Architecture -> Assembly -> Code -> AI
Full-Stack AI Engineer
Zacharia Hammad
I build AI systems end to end: agents, retrieval, computer vision, infrastructure, and the low-level foundations they run on.
Systems depth for production AI
Computer Engineering graduate from Drexel University. I move comfortably from CPU design and virtual machines to production AI pipelines, graph systems, and full-stack product surfaces.
AI systems, from product surface to metal
Four ways to scan the work: product, retrieval, inference, and the systems foundation underneath it.
AI Products & Agents
I build user-facing AI systems that turn models into useful workflows.
Retrieval & Knowledge Systems
I design the graph and vector layers that make AI systems context-aware.
Inference & Edge AI
I ship model pipelines where latency, GPUs, and hardware constraints matter.
Systems Foundation
I understand the lower layers because I have built processors, VMs, and emulators.
Repotoire
Graph-powered code intelligence for AI-assisted engineering.
AI coding agents need accurate codebase context before they can make reliable changes.
Repotoire builds a knowledge graph of a repository, runs language-aware detectors, and packages the result as a fast Rust CLI.
110+ detectors across 9 languages, shipped as a single binary with Homebrew distribution and a GitHub Action path.
Production systems without private details
High-level proof from real work: inference, knowledge systems, product surfaces, infrastructure, and edge deployment.
Built GPU-accelerated video processing paths for production inference workloads.
Designed entity mapping and relationship layers for AI systems that need durable context.
Built dashboard and workflow surfaces that connect AI capabilities to real users.
Maintained deployment paths for production services with repeatable infrastructure workflows.
Deployed model workloads to constrained hardware with hardware-accelerated inference.
"I started at the metal"
Designing CPUs, implementing architectures from first principles. Where I learned how computers actually work.
Custom 32-bit RISC processor with vector extensions in modern C++. Pipelining, hazards, caches, FMA — designed for ML workloads.
Pipelined RISC-V CPU in C with branch prediction, out-of-order execution, and a PC-signature hit predictor. Evolved from single-cycle to fully pipelined.
"Then I built the machines"
Virtual machines, simulators, assembly. The layer between hardware and software.
Rust workspace (ISA / simulator / assembler / RTL) running a nanochat-style LLM on a custom RISC-V target.
16-bit educational architecture implemented in C. Memory-mapped I/O, trap routines; runs 2048 and Rogue.
Python + pygame interpreter for the classic 8-bit virtual machine with full hex-keypad mapping.
"Now I write what runs on them"
Production software, developer tools, and AI systems. Where I am today.
Graph-powered code analysis CLI in pure Rust. 110+ detectors across 9 languages over a knowledge graph of the codebase. Single binary; ships via Homebrew and a GitHub Action.
Rust library for compressed approximate nearest-neighbor search. Implements TurboQuant (Zandieh et al., 2025) on a custom HNSW graph; ~8–14× memory reduction at 4-bit with 0.995+ cosine similarity.
Zero-knowledge IP-layer protocol experiment using STARK proofs, with a Lean 4 formalization and recursive proof support.
The person behind the engineer
Brazilian Jiu-Jitsu
I practice BJJ and help teach the kids' class. The discipline, patience, and mentorship translate directly into how I approach engineering problems.
Chess
My high school chess team won both state and national championships. Chess sharpened my strategic thinking and pattern recognition — skills I use daily in engineering.
Travel & Photography
Exploring different cultures and capturing moments through photography broadens how I think about problems and design.
