Jose Cedeno@geekendzone
2
Events attended
2
Submissions made
United States
10+ years of experience
About me
I am a Senior Reliability Engineer with 30+ years of experience in heavy industry, now building production AI and IIoT systems at the convergence of Operations Technology (OT) and Information Technology (IT). My focus is bridging the gap between physical machinery and data-driven insights through: 🤖 Applied AI / GenAI: Production RAG over technical manuals, LLM-powered tools, and agentic automation embedded in real operational workflows. 🏗️ IIoT Architecture: Prototyping industrial data pipelines from sensor to dashboard. ☸️ Cloud-Native Infrastructure: Orchestrating bare-metal Kubernetes (K3s) for high-availability industrial apps. 📊 Predictive Analytics: Leveraging a Post-Grad in Data Science to turn vibration and sensor data into reliability KPIs. My Philosophy: I don't just monitor assets; I build the infrastructure that makes assets talk.
🤓 Latest Submissions
GeekForge: a dark software factory
GeekForge is a dark software factory: three headless coding-agent seats working together in one BAND Desktop room. The owner sends exactly one message, the dispatch. The planner splits the work and decides every open point itself, the builder implements and commits, and the verifier rebuilds every delivery from a fresh clone in an isolated container with no network, runs its own probes and accepts or rejects with evidence. In the submitted run (tablekeeper track) the band delivered all 4 of 4 stages in 89 minutes from one human message. The verifier ran 308 of its own probes and rejected the first delivery for two defects that no shipped check covered; the builder fixed them and the fix was accepted. The seats also pushed back on each other: the builder refused a mid-stage decision that contradicted the specification, and the planner re-decided. Each seat has its own git identity, a guard hook keeps it inside its own directory, and every brief, decision and verdict is committed. The mandates are generic: to point the factory at a different problem, you change only the dispatch. Full room log, history and measured cost (about 110M tokens, roughly $41 at catalog prices) are in the repository.
6 Oct 2026
DeltaForge
DeltaForge turns a validated momentum signal into an options position an autonomous agent can manage. When price closes above its 55-period SMA on a 30-minute bar — having been under it the bar before, above the 21-SMA, on a bullish candle — the agent buys a slightly-ITM call at 7–14 DTE and lets the underlying's own levels close it: the frozen 8-bar pivot stop, a 3R target, or a 5-DTE clock. The edge was measured before it was traded. Over 2,345 backtested trades on real Alpaca options data, the target is hit only 13% of the time; the edge is the asymmetry at the stop, where a full-R loss in shares costs the call about a third of its debit. The repo also documents the version that failed: the original debit-spread thesis was killed when verification showed its profit rested on five implausible trades. The agent opened every position itself — limit orders at mid, chased no further than the haircut the backtest assumed — journaling every decision to a public dashboard. It finished +7.25% in four sessions, closed to cash on judging day by a coded protective floor, including a measured lesson on that floor's calibration that the write-up does not hide.
4 Sep 2026
