OpenCohost
I built an AI voice co-host for streamers: it ingests Twitch/YouTube chat in real time, generates responses with local LLMs, and synthesizes voice on the fly. Local-first and deterministic.
- Python
- Ollama
- QwenTTS
- Edge-TTS
- Flask
The real transition from improvised code to systems design and agentic automation.
At 24, one week away from receiving my Computer Systems Engineering degree, I hit a wall. The traditional market felt broken: very low salaries, blind recruiting pipelines driven by automated filters, and the threat of professional stagnation. I watched many classmates give up under that frustration and escape into videogames.
It was the worst day because the financial stress and uncertainty were suffocating. I knew I could not compete in a cheap, repetitive labor market. But it was also the best day because I made a hard decision. I uninstalled League of Legends, where I had invested countless hours reaching Emerald rank, and decided that all that mental discipline, focus, and resilience would go into building my own path.
I invested my last five dollars in deepening my web development fundamentals and recording advanced agent-automation methodology webinars. I understood that copy-pasting generated code without understanding Google V8, secure persistence with SQLCipher, or local 12GB GPU CUDA integrations was a trap. That day I stopped acting like a reactive programmer and started thinking like a Systems Architect.
Frontier artificial intelligence did not destroy my career; it amplified it. I quickly understood a core business rule: syntax became cheap, but architecture and distribution are everything. Modern engineering is no longer about manually writing lines of code inefficiently; it is about directing abstract logic.
I adopted Spec-Driven Development and Architecture Decision Records. I stopped programming by blind intuition and started modeling hybrid local and cloud systems. Instead of spending thousands of dollars on closed subscriptions, I integrated agentic orchestration tools such as Codex, Antigravity, and the Engram vector knowledge base under a local-first chassis.
This workflow lets me operate like the director of a technical team inside my own computer. I research, design strategy, audit testing plans, and control the tradeoffs behind every decision — delivering secure, typed, validated software at a speed I would not have imagined as a solo developer.
Systems I designed and built end to end. I use tools like Ollama, Whisper, and Stable Diffusion — but the architecture that orchestrates them is mine.
I built an AI voice co-host for streamers: it ingests Twitch/YouTube chat in real time, generates responses with local LLMs, and synthesizes voice on the fly. Local-first and deterministic.
Live captioning system for streamers: continuous voice detection with Silero VAD, local Whisper transcription, and direct OBS integration. Packaged portably with Nuitka.
Layer-based desktop drawing app, in the spirit of Paint/Photoshop, built with Tauri + Rust. It connects with Sloppy to enhance drawings through local Stable Diffusion XL.
Computer-vision API that generates and improves images with Stable Diffusion XL through local inference. It is the AI engine behind Brick.draw.
Android productivity journal app, local-first, with an encrypted SQLCipher database for privacy. It tracks thoughts, actions, and daily satisfaction.