Technical writeups on LLMs, RAG, VPNs, and self-hosted infrastructure. Posts by Erik Svend Tomberg.
Ecne Project
Filter:
No posts match the selected labels — clear a label to see posts again.
Technical writeups on LLMs, RAG, VPNs, and self-hosted infrastructure. Posts by Erik Svend Tomberg.
A pipeline pulls Vermont legislature data and local news into SQLite; a local 35B model writes a daily digest. Thirteen digest days, a four-month sleep, and two ways the site broke on the way back.
A ~9,700-line Python pipeline turns a keyword into a researched, two-speaker, lip-synced podcast video: persona-YAML prompts, conversational audio padding, sprite-swap lip-sync, and fully local TTS.
A $7 Gear VR controller speaks a custom GATT protocol — reverse-engineered, prototyped in Python, rewritten in Rust, and now a systemd daemon that makes it a working Linux mouse.
Technical breakdown of VPN architecture and cryptography, a Tor comparison, the US/Vermont legal landscape, and a research-backed look at Proton VPN (June 2026 data).
A scheduled two-stage LLM funnel scans ~45 tickers a day, deep-dives the top 3, and paper-trades with dynamic risk thresholds — all on a local 35B model. The architecture, the first days of results, and a full trading day after four months of dormancy.
How I built a citation-grounded RAG chatbot over a 47,000-page wiki: hybrid FAISS + keyword search, domain-specific tagging, and an honest list of where it falls short.
Eight open-source scanners orchestrated into one deduplicated, CI-friendly report — plus a custom SSRF module none of them cover, and what a first full run on a real target taught the design.