isupervillain :: lab

STATUS REPORT
════════════════════════════════════
uptime : 0 days
mode : BUILD >> TALK
agents_active : 1 (MIDAS — in dev)
concepts_queued : 3
markets : CLOSED ○
bullshit_level : 0.00%
════════════════════════════════════
Building AI agents that have actual jobs. 
Right now: one trading copilot (MIDAS) in active development, with three more on the drawing board.
           Nothing shipped yet, but in active build mode.

cat manifesto.txt

001 | MANIFESTO
002 | ────────────────────────────────────────
003 |
004 | Most "AI products" are chatbots with branding.
005 | I care about agents that DO things:
006 |
007 | → Watch market feeds so I don't have to stare at 8 charts
008 | → Flag trade setups before they disappear
009 | → Eventually: monitor livestock sensors, automate ops work
010 | (concept → code → prod — in that order)
011 |
012 | PRIORITIES:
013 |
014 | agentic > scripted
015 | (real autonomy, not if/then flows)
016 |
017 | operators > enterprises
018 | (built for people who ship, not committees)
019 |
020 | shipping > pitching
021 | (code in prod > slides in deck)
022 |
023 | utility > vibes
024 | (does it work > does it look good in a demo)
025 |
026 | EOF

ls -la agents/

AGT-001 :: MIDAS

Status: IN DEVELOPMENT
Role: Day trader's copilot

MIDAS is a research-driven trading agent built to scan, reason about, and act on intraday setups instead of just drawing pretty lines on charts.

Built because staring at TradingView all day is a waste of neurons. Currently running in chunks, frequently breaking, getting better.

Can (so far):

  • Orchestrate sessions with a central clock, kill_switch, and event bus
  • Load rule-based screening configs for repeatable setups
  • Track open positions and session state for post-mortem review

Planned:

  • Pre-market scan to build a conviction-ranked watchlist
  • Multi-agent LLM “analyst desk” per ticker (tech, tape, sentiment, etc.)
  • Decision layer that debates, scores conviction, and chooses LIVE vs ALERT
  • Execution layer to place trades or fire alerts via APIs / bots

Currently cannot:

  • Run the full pipeline end-to-end without supervision
  • Trade fully autonomously with my capital
  • Safely handle every edge-case session without the kill_switch armed

Known failure modes:

  • Breaks when orchestration assumptions change
  • Over-surfacing noisy, low-conviction names from loose screens
  • Gets weird around odd sessions and broken data

[built with: Python + custom agent stack]

AGT-002 :: DEMETER

Status: CONCEPT — NO CODE YET
Role: Agribusiness / livestock monitoring

Livestock health via sensor data. Inspired by SoundTalks. On the backlog. Nothing written yet.

// TODO: find someone with actual farm sensor data

[no repo]

AGT-003 :: PROMETHEUS

Status: CONCEPT — NO CODE YET
Role: Research intel digester

Parse whitepapers, earnings transcripts, output structured notes. Planned. Not started.

[no repo]

AGT-004 :: HEPHAESTUS

Status: CONCEPT — NO CODE YET
Role: Workflow / ops automation

Standup reports, sprint summaries, GitHub issue linking. Planned. Not started.

[no repo]

tail -f lab.log

cat whoami

OPERATOR PROFILE
════════════════════════════════════

Builder, trader, and farmer. Writes code, works the land, and ships weird useful things.

"Supervillain" = outsider mindset. Not trying to look like a hero brand or an
enterprise product. Building from a farm lab, optimizing for edge and utility
over pitch decks.

Day job: software dev. Side projects: autonomous trading tools, ag-tech
experiments, and workflow automation that actually saves time.

STATS/
agents_in_flight    : 1 (MIDAS)
concepts_in_backlog : 3
core_frameworks     : 1 (custom agent stack)
marketing_budget    : $0
built_in_public     : not yet
trading_hours       : market open → market close (and then some)
farming_hours       : excessive

TODO: ~~sleep more~~ add better failure replay harness

EOF

cd contact/

$ ls
brief.txt  reach_me.txt

$ cat brief.txt
Traders who want AI that watches markets, not chats about them
Devs building agentic tools who want to compare notes
Anyone with a boring repetitive workflow that should be automated but isn't yet

$ cat reach_me.txt
github -> https://github.com/isupervillain
email  -> mailto:contact@isupervillain.com

$ curl isupervillain.com/status
{
  "status": "OPERATIONAL",
  "agents": 1,
  "open_to_collab": true,
  "bullshit": 0
}
loading agent_framework.py ████████░░ 80% (still fixing the last 20%)
connecting to trading backend... OK
connecting to humans... OK (signals: traders, devs, farmers)
spawning MIDAS instance... OK (experimental)
DEMETER... CONCEPT ONLY — skipping
PROMETHEUS... CONCEPT ONLY — skipping
HEPHAESTUS... CONCEPT ONLY — skipping
isupervillain.lab ready. (mostly)