Data arbitrage at scale
RuneStock Engine: an analytics platform built on an unregulated commodity market with 400M+ price datapoints.
The Old School RuneScape Grand Exchange is a real-money-equivalent commodity market with continuous price discovery, no circuit breakers, and full information asymmetry. I built an analytics engine to exploit it. The methodology transfers directly to any information-dense commercial environment.
- FastAPI backend, DuckDB (columnar analytics), Streamlit frontend, Docker — production stack, not a notebook
- Live Grand Exchange price feed ingestion via OSRS Wiki API → columnar storage → margin calculation → ranked opportunity output
- Tax/margin analysis module: buy/sell spread, GE tax, net margin per unit, position sizing for liquidity-constrained assets
- Shop arbitrage detection: NPC shop prices vs GE market prices — identifies consistent positive-expectation trades
- Liquidity scoring: volume-weighted opportunity ranking so capital isn't wasted on illiquid positions
- Flipping, proc, and Slayer dashboards — different margin profiles for different risk tolerances and time horizons
Operational finance tooling
Finance dashboard backed by live transaction data — not a spreadsheet.
Built a personal and business finance monitoring system as part of the T-drive MCP gateway: Barclays and Nationwide CSV import pipelines, SQLite finance database, action signals, spend summaries, and a live dashboard payload served via MCP tool to any AI agent in the stack.
- Barclays and Nationwide statement CSV import: normalised transaction schema, deduplication, category tagging
- SQLite finance DB with read-only SQL access via
finance_queryMCP tool — any model in the stack can query spend without touching raw files - Action signals: automated identification of pending decisions, anomalies, and spend trends
- Dashboard payload: JSON summary of income, outgoings, category breakdown, and flagged items — served to monitoring and reporting agents
- Multiple companies tracked: BamPav Ltd, Mars Purifier Ltd, personal — consolidated view across entities
Risk management thinking
Canary traffic controller: the same reasoning pattern as position risk management.
The T-V2 canary controller isn't just a deployment tool — it's a risk management system with position sizing, stop-loss logic, and a hard gate before full exposure. The mental model maps directly to trading and operational risk frameworks.
- Position sizing: v2_weight increments by 10% per passing probe — small initial exposure, scale up on evidence
- Stop-loss: consecutive failure threshold triggers atomic rollback to v2_weight=0 — cut the position before losses compound
- Hard gate:
v2_api_ready=truemust be explicitly set by a human — no automated full exposure, ever - Full audit trail in
traffic_split_history: every weight change, trigger source, and rollback recorded — position log - Spend discipline baked in: LLM usage tracked via
llm_usage_ledgerwithcost_micro_usd BIGINT— every token has a price