Curriculum Vitae

2025.07 – present

Quantitative Analytics Associate

Company: Wells Fargo

• Selected into a competitive rotational development program rotating across lines of business, with formal training in banking, finance, and regulation.
• WIM Model Risk (Jan 2026 – Jul 2026): Validated the tracking-error prediction model used for risk projection across Wells Fargo wealth-management client portfolios, benchmarked against BlackRock Aladdin.
• Implemented and compared multiple factor models (Fama-French 5-factor, a 12-factor model, full sample covariance, experimental PCA factors) from research papers, with research-based extensions focused on covariance and volatility scaling — market-cap covariance scaling, EWMA weighting, and Barra-style GARCH/GJR-GARCH and MF2-GARCH volatility scaling — that improved tracking-error accuracy.
• Built a full backtesting engine supporting fixed and varying benchmark/portfolio configurations, plus a Dirichlet-based testing-portfolio construction method that holds portfolios market-cap weighted on average while modeling realistic variance across the client base.
• Risk Modeling Group, Auto Decisioning (Jul 2025 – Dec 2025): Built and shipped to production the second-stage Eagle model, a probability-of-charge-off model that makes the final approve/decline decision on all Wells Fargo auto originations.
• Handled variable selection for fairness compliance, monotonicity validation against drivers like FICO, and segment-level error analysis on realized vs. predicted charge-off.
• Separately built an automated auditing tool for credit-bureau attributes, replacing a manual process that required a team of ten analysts working multiple weeks.
• The tool parses the attribute specification sheet, applies its rules to applicant credit histories, and reconciles against the data provider, raising validated coverage from under 1% of applicants to 25%+ in a single day.

2025.01 – 2025.05

AI Engineer (Contract)

Company: Seyon Solutions

• Architected and built from scratch an agentic LlamaIndex assistant for a chemical-processing client operating two plants (650+ employees), letting managers query worker performance and plant operations in natural language.
• Designed a multi-tool agent: a primary few-shot SQL tool (~100 pre-written CTEs stored with natural-language descriptions in a vector database, similarity-matched at query time and passed as few-shot examples to a SQL-writing LLM with schema context), an AWS OpenSearch vector-search fallback, and a graphing tool that produced charts directly from natural-language queries.
• Designed and implemented a bronze-silver-gold medallion pipeline on AWS: append-only event JSON landed via Lambda into a terabyte-scale bronze store, a Lambda-triggered silver layer maintaining active state and full history via Delta Lake, and a Glue job aggregating to production gold tables every five minutes.
• Deployed to real users with ~90% query accuracy by end of engagement.
• Stack: LlamaIndex, AWS OpenSearch, Streamlit, AWS Lambda, AWS Glue, Delta Lake, SQL.

2024.08 – 2025.01

AI Solutions Consultant (Contract)

Company: Chartwell Insurance Services

GitHub: ZachFara/Chartwell-Insurance-AI

• Built a production RAG-powered customer-service assistant for Chartwell's support team.
• The system ingested insurance policy documents, chunked and indexed them into Pinecone for semantic search, and retrieved context at query time to generate professional email-formatted responses.
• Built a dedicated hyperparameter-tuning harness that swept chunk size, overlap, top-k, and prompt variations against a question dataset with results logged across iterations, and experimented with alternative chunking strategies.
• Modular architecture separating agent orchestration, vector-store management, document loading, and configuration, with live document upload to keep the knowledge base current.
• Stack: GPT-4, Pinecone, Streamlit, LlamaParse.

2024.06 – 2024.08

Quantitative Analytics Intern

Company: Wells Fargo

• Worked on Fargo, Wells Fargo's production AI virtual assistant in the consumer banking app (245M+ interactions in 2024 across 33M+ mobile active users).
• Designed and ran a demographic bias-evaluation suite for Fargo's intent-classification model, systematically varying gendered pronouns and ethnicity references across hundreds of queries to measure output consistency, including edge cases where demographic context is legitimately relevant.
• Measured 90%+ classification consistency across demographic variations.

2023.06 – 2024.06

Data Scientist, Engineer I

Company: Badger Meter Inc.

• Built a large-scale AWS EMR + Spark ETL pipeline aggregating 5 TB of structured and unstructured telemetry from DynamoDB into a central warehouse, optimized with multithreading, parallel computing, and C++/Python interoperability.
• Engineered a gradient-boosting classifier to predict device failure from voltage telemetry on a highly imbalanced dataset with few labeled failures (~60% F1, a strong result given the signal and label constraints), plus a time-series regression model for voltage-degradation patterns.
• The model formed the basis of a proposed paid device-health API for out-of-warranty devices.
• Served as the sole quantitative source of truth for fleet health across a 1,000+ device pilot (path to a 3M+ device fleet), delivering monthly diagnostics to senior leadership.
• Stack: AWS EMR, Spark, DynamoDB, Python, C++, XGBoost, time-series modeling.

2020.08 – 2023.05

Data Science Researcher

Company: Marquette University

• Joined a CS research lab alongside PhD students (partially funded by NSF grant #1950826).
• First-authored and co-authored back-to-back IEEE Big Data publications (2022 and 2021) on hospital length-of-stay prediction from electronic health records, owning the full pipeline on the 2022 paper: data processing, modeling, Bayesian hyperparameter sweep, results, and analysis.