SIGNALS
OVER NOISE.
I'm a computer-science grad (CSUF) who spends his days teaching STEM to elementary-school kids. This site is my night-shift: a long-running experiment in whether LLM agents can actually trade — and whether the work can be published in public, every fill and every flop.
Eight strategies score every optionable ticker before the bell. A quant scanner ranks them. Alpaca pulls the trigger. A team of specialist agents watches the model around the clock. If it makes money with my small bankroll, that's cool. If it doesn't, I'll learn exactly why — in public.
Curious by default. Optimistic about the future. Publishing the work so others can learn alongside me.
Heads up — this is an experiment. The bankroll is small, the model is iterating, not every session ends green. None of it is financial advice. The point is to run the thing in the open and let you watch it improve.
Pre-market — eight strategies score every optionable ticker. Thousands of candidates feed through the pipeline.
Confluence scoring collapses the field to a ranked ten. Edge, confidence, skew, liquidity, news attribution.
Alpaca fires the book at open. Auto-trade armed on a real bankroll with hard stop + daily loss cap.
Every fill, every close, every P&L — live to Discord + the public ledger. No edits, no selective memory.
Five layers. Market data comes in, edge is computed, gates filter out noise, Alpaca sizes the contract into one of three risk buckets, and every outcome posts to the tape. Nothing is hidden.
Minasara orchestrates. Six specialist agents own a slice of the stack each, run on timeslots, and post findings to Discord channels the user can read. Agents start each task fresh — their memory lives in versioned context files, not conversation history.
Product manager for the whole stack. Assigns tasks to the specialists, handles user comms, owns git + deploy. Runs around the clock on a home server.
Competitive intel, academic papers, sector research, UI/UX direction. Rotating deep-research dig every night.
P&L attribution, win-rate by strategy, risk metrics, calibration tracking, unit economics.
Graduated. Curious about agentic systems and whether LLMs could make decisions traders make under uncertainty.
Single-strategy scripts, Jupyter notebooks, paper trades. Discord started with a small group of friends.
Six specialist agents, a scanner with eight strategies, confluence scoring. VPS deployed on OCI. First public ledger entry.
Auto-trade goes hot on a real bankroll. elevan.vzn opens publicly. Track record starts its tape in real time.
When the model proves profitable over months — paid seats, live signal feed, community discord. Until then: waitlist + viewer access + learn in public.