Profile

Reproducible red-team + AI risk assessment of a tool-using LLM customer-service assistant (OWASP LLM Top 10, MITRE ATLAS, NIST AI RMF). Synthetic target

LLM pipeline: SEC 10-K risk-factor extraction, 14-class classification and YoY change detection, scored against a hand-labeled gold set

PostgreSQL schema + 15 analytical queries (CLV, RFM churn, cross-sell, cohorts) with a DuckDB pipeline and measured EXPLAIN ANALYZE tuning. Synthetic data.

Notes and sanitized configs from running a self-hosted lab: Cloudflare Tunnel (no inbound ports), Authentik SSO, local OpenAI-compatible LLM, stdlib health checks.

RAG Q&A over real FEMA flood insurance policies with clause-level citations, refusal handling and a 120-question evaluation harness

MCP server giving Claude Code 9 tools to message, assign tasks to, and share context with agents on other machines through a REST hub (hub not included). 13 tests.

Daemon that joins an agentchattr hub over WebSocket and answers @mentions by running Claude Code locally or over SSH. Loop guard, sender allowlist, 18 tests.

11 Power BI DAX measures (budget variance, MoM, running totals) reproduced in DuckDB SQL with tests, plus anomaly detection and Holt forecast. Synthetic data; original .pbix lost.

Self-healing monitor for a Docker homelab — evidence collection, heuristic + local-LLM diagnosis, policy-gated auto-restart, markdown postmortems. Zero runtime dependencies.

Insurance portfolio analytics on synthetic data (68K transactions): Excel pivots + Python, Welch/bootstrap/chi-squared tests, hurricane Monte Carlo PMLs and XoL layer sizing.

Motor insurance pricing (Poisson/Gamma GLM vs LightGBM) on 669k real policies, with an SR 11-7/26-2-style validation and NAIC/NIST governance pack

dbt + DuckDB star schema over FEMA's 2.7M NFIP flood claims: earned exposure, loss trends, severity model, dashboard, underwriter memo

FAIR-style Monte Carlo cyber risk quantification from insurance-application controls: readiness score, dollar-ranked control gaps, Excel workbook. Hypothetical firms

End-to-end IT general controls (ITGC) audit of a lab AD/PostgreSQL/git environment: code-driven evidence, sampling, workpapers, management letter. Synthetic company

Weighted auto insurance risk scoring engine — 8 actuarial factors, tier + premium multiplier output

Patient portal from a UGA MIST 4610 group project (Node/Express/SQLite): enforced TOTP 2FA, RBAC + per-patient access checks, hash-chained audit log, 55 tests, NIST 800-53 / HIPAA 164.312 control m…

Monte Carlo captive feasibility study for a hypothetical $50M manufacturer: GC vs large deductible vs captive, domicile comparison, Excel workbook + CFO memo. Simulated loss data

Autonomous Kalshi weather-market trading bot — 31-member GFS ensemble, rolling NWS bias correction, quarter-Kelly sizing, 98 tests. Phase 1 paper-validated.

Push-to-talk voice dictation for Windows — Groq Whisper + Llama 3.1 cleanup, VAD-fired partial transcriptions for low latency, 42 tests

UGA MIST 4610 group project — Tableau analysis of LA crime data

UGA MIST 4610 relational database project