AI/ML engineer and full-stack developer. CS student who fine-tunes LLMs, builds agentic automation systems, and ships production web + mobile apps.
- Won Champion at Hack The East 2026 with Origraph — AI content provenance platform
- Won an MLH track prize at UWB Hacks with Vellum — invisible watermarking for LLM outputs
- Building Umbra — private-prompt inference marketplace on attested Apple Silicon
- Co-founder at Reunify Labs — building FPAI (first-principles AI ed-tech)
- Fine-tuned LLMs (Nemotron, Qwen, GPT-OSS) and DiT audio models on H100s
- 12 models + 8 datasets on HuggingFace
| Area | Details |
|---|---|
| AI/ML | LLM fine-tuning (GRPO/RL, LoRA), DiT audio model training, data collection & curation, PyTorch, transformers |
| Agentic AI | Autonomous workflows (Claude Code, Appium), tool-use patterns, LLM API orchestration, multi-agent systems |
| Full Stack | Python, TypeScript, Node.js, FastAPI, Next.js, Supabase, PostgreSQL |
| Deployment | Vercel, Docker, AWS (Lambda, S3, CloudFront, DynamoDB), Railway, CI/CD, GitHub Actions |
| Data | Dataset creation & curation (HuggingFace), preference datasets, synthetic data pipelines |
| Mobile | Some iOS experience (SwiftUI, App Store deployment) — primarily AI-assisted development |
| Hardware | Arduino, ESP32 — timing control, voltage switching, sensor I/O |
| Project | Description | Stack |
|---|---|---|
| Umbra | Private-prompt inference marketplace on attested Apple Silicon — uncensored + community fine-tunes you can't get from incumbents. | Go llama.cpp Apple Silicon |
| Origraph | AI content provenance — invisible text watermarking SDK, provenance registry, browser extension. Champion at Hack The East 2026. | Python JS Solana |
| Vellum | Enhanced Origraph rebuild — invisible watermarking, on-chain anchoring, provenance verification for LLM outputs. MLH track prize at UWB Hacks. | TypeScript Auth0 Solana |
| FPAI | First Principle AI — ed-tech platform that explains concepts from first principles using recursive block-based learning. Co-founder at Reunify Labs. | Next.js AWS DynamoDB |
| Amadeus 2.0 | Vision-first Android automation — AI-directed app workflows via Appium + Gemini. GUI dashboard, TUI operator mode, scheduled jobs. | Python Appium Gemini |
| ContextCLI | Universal CLI profile launcher — run any dev CLI under a named auth profile. One machine, multiple accounts, zero friction. | Rust Keychain |
| owenisas.com | Personal portfolio as an interactive macOS desktop simulator — 3D desk showroom, working apps, deep linking. | React Three.js Vite |
| Amadeus | Open-source mobile AI assistant framework. Foundation for 2.0. | Python Swift |
Live at owenisas.com — boots into a 3D desk showroom, then zooms into a fully interactive macOS desktop
| Type | Notable | Stats |
|---|---|---|
| Models | gpt-oss-20b-union-identity (21B LLM fine-tune), nemotron-3-nano-reasoning (32B), qwen-3-nano-reasoning |
12 public models |
| Datasets | opus46-reasoning-mix-full, big-math-rl-30k-subset, first-principles-thinking |
8 datasets |
