Projects
My applied projects focus on model validation, data governance, quantitative risk, and AI model risk. The goal is to translate research and quantitative methods into reproducible, employer-facing work with clear validation logic, documentation, and controls.
Independent Validation of a GARCH Volatility Model
Focus: Market Risk · Model Validation · VaR Backtesting
Independent validation of a GARCH(1,1) volatility model using S&P 500 and EUR/USD daily returns.
Key components:
- Conceptual soundness and parameter diagnostics
- Independent re-implementation of the GARCH likelihood
- Benchmarking against EWMA, Student-t GARCH, GJR-GARCH, EGARCH, and Historical Simulation
- 95% and 99% one-day VaR backtesting
- Kupiec and Christoffersen coverage tests
- Basel traffic-light analysis
- Parameter stability and stress-period analysis
- Optional Black-Scholes pricing extension to study volatility-model risk transmission
Skills: Python, time-series econometrics, VaR, Expected Shortfall, model benchmarking, implementation verification, sensitivity analysis
Status: In development
Point-in-Time Financial Data Governance Pipeline
Focus: Data Risk · Backtesting Infrastructure · AWS
A point-in-time financial data pipeline designed to prevent look-ahead bias and reconstruct the information actually available at each historical forecast origin.
Key components:
- FRED / ALFRED real-time vintage data
- PostgreSQL point-in-time database
- Release-date-aware joins
- Publication-lag and revision analysis
- Automated schema, integrity, coverage, and outlier checks
- Comparison of final-vintage versus historically available data
- AWS deployment using S3, RDS PostgreSQL, Lambda, EventBridge, and CloudWatch
- Least-privilege IAM and reproducible data lineage
Skills: SQL, PostgreSQL, AWS, data engineering, data governance, look-ahead controls, model-input validation
Status: In development
Independent Validation of a Retrieval-Augmented LLM System
Focus: AI Model Validation · RAG · Model Governance
A validation framework for a retrieval-augmented generation system built on public regulatory and model-risk documents.
Key components:
- Manually verified evaluation set with in-scope, out-of-scope, and adversarial questions
- Retrieval evaluation using Recall@k and MRR
- Generation evaluation using faithfulness, citation accuracy, answer relevance, and hallucination rates
- Prompt-sensitivity and non-determinism testing
- Chunk-size and context-ordering sensitivity
- Failure taxonomy separating retrieval, generation, and corpus failures
- Self-hosted RAG system compared with a managed vendor implementation
- Model Card, Validation Report, and Monitoring Plan
Skills: LLM evaluation, RAG, AI model risk, Python, AWS Bedrock, S3 Vectors, model governance
Status: In development
Model Monitoring and Inventory Control Center
Focus: Ongoing Monitoring · Model Governance · Controls
An integration layer that consolidates monitoring outputs from multiple models into a model inventory and control framework.
Key components:
- Model inventory and tiering
- Validation-cycle tracking
- Input drift monitoring using PSI, KS, and Jensen-Shannon divergence
- Rolling performance monitoring
- Green / yellow / red escalation rules
- Issue and remediation tracking
- CloudWatch metrics and alarms
- Automated management reporting
Skills: Model monitoring, governance, drift detection, issue management, AWS CloudWatch, control design
Status: Planned integration layer
Credit Capital Model Validation
Focus: Credit Risk · Regulatory Capital · Monte Carlo
Validation of the Vasicek / ASRF credit-capital framework, including finite-portfolio effects, parameter sensitivity, and granularity risk.
Key components:
- Closed-form ASRF capital calculation
- Monte Carlo validation
- Portfolio granularity analysis
- PD, LGD, and asset-correlation sensitivity
- Procyclicality and stress analysis
Skills: Credit risk, Basel capital, Monte Carlo simulation, sensitivity analysis
Status: In development