I ship end to end across Go, C++, TypeScript and Python, from GPU training infrastructure and self-training AI agents to real-time apps at the edge.
Most recently: an on-device lip-reading pipeline at StormHacks, a loop that turns an agent's repeated work into small fine-tuned specialists, and a summer as a Technical Content Engineer at LicenseSpring writing SDK samples and developer education. Final-year CS at Langara. Daily driver: Arch.
now
building the Langara CS Club website redesign
building a universal post-secondary Discord bot
studying final year of CS at Langara
-
heard
StormHacks 2026 · team of 4
Silent-speech app: reads your lips from a webcam and speaks for you. I built the on-device lip-reading pipeline (Auto-AVSR → ONNX, int8, 775 MB → 203 MB, in the browser) and the FastAPI GPU inference service on RunPod. · devpost -
hermes-apprentice
winner · Hermes Agent Challenge
A second learning loop for Hermes Agent: distills its recurring work into 18 MB QLoRA specialists on vLLM (~38 ms p50 local vs multi-second API calls), each gated by a held-out F1 test and a canary ramp. · writeup · devpost
- Your AI slop bores me · Jun 2026
- Skills are Prompts. Here's how Hermes Apprentice turns them into weights · May 2026
languages Go · C++ · Python · TypeScript · SQL · Bash
backend Hono · FastAPI · Next.js · Drizzle · PostgreSQL
infra Docker · Kubernetes · Proxmox · Firecracker · Cloudflare Workers/D1 · GitHub Actions
ml / ai LoRA fine-tuning (Unsloth) · local LLMs · MCP servers · RAG (Vectorize)
homelab 3 Proxmox nodes, live status on eschmechel.dev/~
eschmechel.dev · linkedin · dev.to
_ /\=/\
(_) (=o.o=)
/|_|\___________________ / \
| | \|_______|
_| |_ (o) (o)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~