I build things. Web apps, ML pipelines, Android automation tools, production APIs — if it can be built, I'll build it. Speech models fine-tuned in PyTorch. Platforms shipped solo. Pipelines running on nothing but a phone. Full-stack, end-to-end, whatever it takes.

End-to-end TikTok automation pipeline running entirely on a phone via Termux. Watches a link inbox, downloads videos in HD, extracts frames with ffmpeg, and generates captions via GPT-4o Vision. Async 5-worker pipeline with parallel captioning across 3 concurrent workers. FastAPI control layer for job orchestration, retry logic, and per-worker failure isolation.

Built because constraints are just problems that haven't been engineered yet.

Impact: ~58% faster content processing (2 hrs → 50 min for 40 videos), fully hands-off.

Python Kotlin Android FastAPI OpenAI API ffmpeg Termux

Visual, node-based engine for building and validating AI pipelines with real-time execution. 9 modular node types (LLM, API Call, Vector Store, Conditional, etc.) on a reusable BaseNode architecture. Regex-driven variable parsing ({{ variables }}) auto-generates graph connections. FastAPI backend validates DAG integrity via Kahn's Algorithm before execution.

Turned a static ReactFlow demo into a working visual programming system.

React ReactFlow FastAPI Python DAG Algorithms

Fine-tuned a 24-layer Transformer for 8-class emotion recognition across RAVDESS, IEMOCAP, and CREMA-D. 77.3% accuracy in-distribution; cross-corpus testing exposed a 16–23pp generalisation drop, diagnosed via arousal/valence failure analysis — a known limit of self-supervised speech embeddings. Scoped fixes: adversarial domain adaptation, speaker disentanglement, Kenya-specific DAPT corpus.

Full training-to-failure-analysis pipeline — not just a trained model.

PyTorch HuggingFace WavLM SpeechBrain Transfer Learning

Scalable B2B communication platform helping businesses engage customers at scale. Shipped under competitive hackathon pressure with a 2-person team, full-stack, won against the full field.

Deceptively simple-looking. Actually not.

Scalable Architecture B2B Rapid Deployment Full-Stack

Three production platforms — full lifecycle ownership, architecture to deployment to support.

Next.js React Node.js REST APIs Vercel SEO

Machine Learning Engineer Intern · JHUB Africa  ·  May – Aug 2024

• Integrated ML models into production systems (React + FastAPI) — owned the model-to-UI connection end to end.
• Designed RESTful APIs bridging backend inference pipelines with frontend components.
• Code reviews and technical presentations to non-technical audiences in async, remote-adjacent teams.

Full-Stack Developer · Independent Clients (Freelance)  ·  2024 – 2025

• Architected, deployed, and maintained 3 production web applications solo, 100% on-time delivery.
• Owned the full workflow: REST API design, version control, Vercel deployment, SEO, post-launch support.

BSc Computer Science — Jomo Kenyatta University of Agriculture & Technology (Expected 2026)

Data Structures & Algorithms · Artificial Intelligence · Operating Systems · System Design · OOP · Cryptography · Probability & Statistics · Internet Application Programming