Looking forward to building something that truly matters.
Discipline drives me. Design defines me. Humanity guides me.
| Project | Description | Stack |
|---|---|---|
| Eve's Eye | Real-time AI security monitoring. Gemma 4 31B on Cerebras Cloud for threat detection. Built at the Cerebras x DeepMind hackathon. | Next.js 16, Cerebras SDK, SQLite |
| camx2 | Local-first AI surveillance system. ONNX + WebGPU, no cloud dependency. Multi-camera detection, VLM threat verification, video Q&A pipeline. | Next.js, ONNX Runtime, WebGPU |
| muxics | Winamp-inspired cross-platform desktop music player and unofficial YouTube client. Works everywhere: macOS, Windows, Linux. | Electron, React, Vite |
| bukmarks | Bookmark dashboard that imports Chromium bookmarks into a clean, searchable, animated interface. Browser extensions for Chrome and Firefox. | Next.js, Convex, Better Auth, GSAP |
| yotes | Privacy-first notes app. Client-side AES-GCM encryption, real-time sync, PWA, AI-powered search summaries via Gemini. | React, Convex, Gemini |
| patchcore_service | Package damage detection service using PatchCore anomaly detection. Web UI and REST API with CORS. WideResNet-50-2 backbone. | Python, Flask, PyTorch |
| avdbuddy-win | Desktop app for managing Android Virtual Devices without Android Studio. Guided creation wizard for all form factors. | Bun, webview, TypeScript |
| Your Universe | An X-like social platform with chat, feed ranking, research paper search, and scholar discovery. Live demo (read-only). | Next.js |
Published on Hugging Face:
| Model | Description | Type |
|---|---|---|
| Chinchilla-1-73M | Research-scale 73.5M parameter instruction-tuned LM. Custom nanochat architecture with value embeddings, smear mechanism, backout residuals. Trained on Modal GPUs (~$1.20). | Text Generation |
| LFM2.5-VL-1.6B UCF Crime (GGUF) | Surveillance crime detection model fine-tuned on UCF Crime dataset (15 categories). LoRA adapters + GGUF quantized. +9.6% accuracy over base model. | Image-to-Text |
| LFM2.5-VL-400M UCF Crime (GGUF) | Lightweight 400M variant of the UCF Crime detection model. LoRA adapters + GGUF quantized. Trained on free Colab T4 GPU. | Image-to-Text |
| inventorygemma | Grocery inventory management expert, fine-tuned from Google Gemma-3-270m. Analyzes stock levels and generates restocking recommendations. | Text Generation |
I Fine-Tuned a Vision Model for Crime Detection on a Free GPU — Mar 2026
How I Automated My Jellyfin Music Library Cleanup with Python — Oct 2024
Yotes: My Quest to Build a Notes App That's Actually Mine — Feb 2024
Daily — TypeScript, JavaScript, Python · Next.js, Tailwind CSS, Shadcn/ui · Convex, Supabase, Prisma, Better Auth, SQLite · Electron, React Native
AI — ONNX Runtime (WebGPU), YOLO, llama-cpp, Unsloth, Transformers.js, Cerebras Cloud SDK, Google Gemini
Also — Node.js, Flask, Astro, Vite, GSAP, Framer Motion · Bun (webview), Win32 API · Java, C, C++, Solidity, Elixir, Shell · Docker, Vercel, Cloudflare, Modal, GitHub Actions · Linux, Git, Biome, Figma, pnpm