Abdelrahman Khaled

I’m a full-stack TypeScript developer and Computer Science graduate (3.84/4.0 CGPA) focused on building production web applications and applied AI systems. I like shipping end-to-end features—from data models to secure authorization and background workflows—while keeping performance and correctness in mind.

Most recently, as the sole developer on Automart, I built a dealership operations platform covering staff, sessions, customers, vehicles, inventory, permissions, reports, and AI-assisted workflows. I maintained 662 passing Vitest tests across 168 test files and kept a clean TypeScript check (tsc --noEmit), and I engineered a hybrid authorization system with a 65-permission catalog, AES-GCM credential encryption, and audit logging. I also built a streaming AI assistant with sandboxed, permission-aware SQL reporting, Redis caching with timeouts and a circuit breaker, and cross-instance PostgreSQL coordination for Redis failover.

In Mitos, I’m developing an AI developer tool that turns natural-language questions about GitHub repositories into commit-pinned execution traces with synchronized Monaco code views, Mermaid diagrams, and navigable timelines. I designed a two-phase AI pipeline for repository exploration and structured trace synthesis, backed by symbol indexing, deterministic verification, quality scoring, and conditional self-correction. I orchestrated durable Inngest workflows with attempt-scoped idempotency, Redis worker locks, SSE status streaming with polling fallback, trace-scoped access controls, and AES-256-GCM encryption of GitHub OAuth tokens.

Alongside these, I built abdok.dev—a unified full-stack platform for case-study portfolio publishing and a repository-aware photography system with hybrid photo discovery (Vertex AI embeddings + PostgreSQL full-text search + Gemini reranking), timezone-aware spotlights, and an immersive 3D gallery using React Three Fiber. For research, my AutoCompile project focuses on LLM-based LLVM optimization pass-sequence prediction (LLVM 19, PyTorch, Hugging Face Transformers), where I co-developed a tool-free pipeline that avoids iterative compiler feedback during one-shot inference. I’ve also earned recognition like 2nd Place at SolveTHE17 and multiple hackathon finalist placements, which reinforces my drive to deliver useful, real-world systems.