Maya Buto@mayasystemslab
2
Events attended
2
Submissions made
United States
1 year of experience
About me
AI systems architect and developer building custom AI agents, workflow automations, and end-to-end business systems. I work with LLMs, n8n, APIs, Telegram, Instagram, Bitrix24, Google Sheets, AppSheet, and human-in-the-loop workflows. My portfolio includes AI lead qualification and CRM routing, support QA analytics, a stateful beauty AI agent, Telegram channel benchmarking, and a paid digital measurement and pricing system for a construction company. I focus on reliable AI systems with structured data, deterministic business rules, human approval, and controlled business processes.
🤓 Latest Submissions
Market Jury — AI Investment Council
Market Jury is a safety-first AI investment decision system built for the Alpaca AI Trading Agents Hackathon. Instead of letting one model reason and trade in the same path, Market Jury separates analysis authority, decision authority, capital authority, and broker execution. The public web app reads live Alpaca IEX market data and an Alpaca Paper account, while persisted research evidence is clearly distinguished from live prices. The system surfaces Bull, Bear, Red Team, and independent Judge reasoning so competing arguments and unresolved evidence gaps remain visible. In the current real demo, the latest governed decision is WATCH. Material evidence gaps remain unresolved, so capital is not released, human trade approval is not reached, and no Alpaca order is created. Market Jury also includes a zero-call Initial cost preflight. Before any future paid model execution, it freezes the exact request set, hashes it, calculates the maximum possible cost, requires explicit owner approval, and disables automatic paid retries. Cost approval does not grant trade authority. Portfolio, Trades, and History make broker capacity, authority state, and decision lifecycle observable. A prior INVEST decision expired under the time-to-live policy before a fresh WATCH decision, preventing stale authority from continuing to move capital. The public demo is built with Python 3.12, FastAPI, Jinja2, JavaScript, Uvicorn, and Pydantic, and is deployed on Render. Alpaca access is Paper-only. Broker writes = 0, Alpaca orders = 0, public-demo model calls = 0, and live money is prohibited. Core principle: Analysis authority ≠ capital authority ≠ execution authority.
4 Sep 2026
FieldOps AI
FieldOps AI is a Release 1 prototype for managers at small construction and renovation companies. It converts a synthetic field sheet, sketch, and voice-note source into an evidence-linked typed Job Graph. native.builder was used to design and iterate the five-screen public workflow; the final web build is deployed on Netlify and requires no login. A live OpenRouter GPT-4o path interprets the field sheet and sketch together with the note transcript, extracting candidate zones, dimensions, openings, work items, and provenance. AI does not approve the package or calculate prices. If a critical physical fact is missing, the system abstains: sill quantity stays UNKNOWN, the status is REVIEW REQUIRED, and release is blocked until a human confirms it. After confirmation, fixed deterministic rules calculate geometry, quantities, client pricing, and installer wages. The same approved Job Graph creates consistent Client, Materials, and Crew views. In the verified golden path, the system keeps active 3000 mm separate from crossed-out 2900 mm, receives human confirmation of 2 sills, and completes the live end-to-end flow in 2:29. Release 1 uses synthetic data only and makes no claims of production-ready procurement, general multilingual support, measured savings, or real-client deployment.
10 Aug 2026
