I turn ambiguous operational problems into practical data and AI systems—connecting stakeholder discovery, solution architecture, implementation, evaluation, and adoption.
One candid note: this GitHub is deliberately AI-polished. I use AI tools to help structure documentation, accelerate selected scaffolding, and pressure-test how the work is presented. The builds, datasets, experiments, ambitions, decisions, and results are real. I review, test, and own what I publish, and I state maturity and limitations where they matter.
Portfolio · Data Science · Analytics · Applied AI · Responsible AI
- Building citation-grounded knowledge and document-intelligence systems.
- Designing evaluation gates, human review, audit evidence, and monitoring for high-trust AI workflows.
- Applying data science, analytics engineering, and solution architecture to public-sector, healthcare, operations, finance, and real-estate problems.
- Targeting forward-deployed, applied AI, enterprise solutions, and AI transformation roles.
Status: Completed analysis. Evaluated 101,766 real de-identified encounters with leakage control, calibration, subgroup errors, SHAP, and a model card. A measured 0.44 age-group false-negative-rate gap supports the documented recommendation not to deploy as-is.
Status: Completed analysis. Designed a config-driven benchmark of five pipelines across three real domains and random/shifted splits. Preprocessing added up to 0.19 AUC while tuning and AutoML added almost no mean lift under the tested conditions.
Status: Completed analysis. Built a DuckDB star schema, data-quality gates, supplier-risk logic, and five-page Looker Studio decision flow over 98,666 real orders. Late orders were 6.5× more likely to receive a 1–2-star review.
Status: Deployed demo. Built a citation-grounded RAG workflow that gives multiple models the same retrieved evidence, separating retrieval quality from generation behavior. Includes ingestion, OCR fallback, FAISS retrieval, evaluation, FastAPI, Docker, and a live demo.
- Data and analytics: Python, SQL, statistical analysis, KPI design, dimensional modeling, BI, and geospatial analysis.
- Machine learning: calibration, feature engineering, subgroup analysis, SHAP, drift evaluation, and temporal validation.
- Applied AI: RAG, embeddings, vector retrieval, document ingestion, OCR, FastAPI, Docker, grounding, and citation evaluation.
- Responsible AI: fairness evaluation, model cards, human review, auditability, risk registers, and deployment gates.
- Solution delivery: stakeholder discovery, problem framing, architecture, prototyping, implementation planning, and technical communication.
Discover → Architect → Implement → Evaluate → Adopt
I start with the user, workflow, value, data, and constraints. I then design and build the core system, test quality and failure modes, and translate the result into a responsible deployment or do-not-deploy decision.
My longer-term trajectory is technical product building and venture exploration grounded in data science and AI. Check out Product and Venture Lab if interested. It is organized around one thesis: AI becomes genuinely useful when context, authority, evidence, and human decision-making are connected. Aegis is the current independent product direction; StepLens is the longer-term interface opportunity. The remaining explorations are preserved as an honestly staged archive, not presented as simultaneous companies.
Some blabberring pieces over at Blog