Building intelligent software that transforms data into decisions.

A multi-agent AI system that simulates a football club's operations department, evaluating players through specialized reasoning agents instead of one model trying to do everything at once.

Problem

Player evaluation blends statistical performance, tactical fit, and financial feasibility. Asking a single model to reason across all of it at once produces muddled analysis, and gives the model no reason not to fill data gaps with confident guesses.

Solution

Specialist agents — Scout, Performance Analytics, Tactical Fit, Transfer Market, and a Devil's Advocate — each own one slice of the analysis and communicate through shared, schema-enforced contracts. Every agent is grounded in an actual player data layer and stays explicit about what it can't verify rather than filling gaps with model recall. A General Manager agent orchestrates the specialists; a Report agent turns their findings into a narrative without being allowed to redecide the verdict.

Outcome

A single evidence-based recommendation — Buy, Monitor, or Do Not Sign — with a full audit trail of sources used, data freshness, and a built-in challenge to the majority view, surfaced through an interactive Streamlit interface.

A financial research platform that keeps a tracked list of companies current automatically, built for long-term investment reasoning rather than headline-chasing.

Problem

Long-term research decays fast: financials go stale after earnings, relevant news gets buried in noise, and re-reading filings for a dozen companies every quarter doesn't scale by hand.

Solution

A CLI-driven automation layer refreshes financials quarterly and news daily per ticker, pulling from Yahoo Finance and generating AI business analysis through a provider-agnostic client that switches between Anthropic and OpenAI with a single config change. Every ticker is processed independently, so one failure never aborts the run for the rest. Results are written into a structured, versioned Excel workbook.

Outcome

A research file that stays current on its own, cross-references companies against each other, and reasons from the underlying numbers instead of restating consensus opinion.

Reconciliation automation built during my internship at Investcorp, comparing bank-level financial records across independent systems that had previously been matched by hand.

Problem

Reconciling two independent record sets — the fund accounting system and bank statements — by hand is slow, and mismatches are easy to miss until they compound.

Solution

A pipeline matches transactions exactly first, falls back to fuzzy matching for near-duplicates the exact pass misses, then classifies whatever's left into labeled exception categories with a confidence score instead of leaving them as an undifferentiated pile.

Outcome

Cut manual review time by surfacing only the transactions that genuinely need a human look, with every match traceable back to its source records.

An interactive dashboard tracking technology spending across business units, built to replace a static monthly reporting cycle.

Problem

Leadership needed a current view of technology spend across business units without waiting on the next manual report.

Solution

Built interactive dashboards using DAX measures, slicers, and drill-through functionality, connected directly to live financial databases so figures reflect same-day activity rather than a monthly export.

Outcome

Replaced static monthly decks with a self-serve tool finance and line-of-business leadership could query directly in planning meetings.