Projects
Talcott Resolution 2023 – Present
GEAR — Enterprise Reporting Platform ACE Award Winner · 2026
Actuarial and investment reporting ran on manual data extraction, heavy Excel manipulation, and email distribution. It was slow, hard to scale, and exposed to human error and key-person dependency — and some analysis was never produced at all, because there was no capacity to assemble it by hand.
I built GEAR (General Email and Reporting Application), a cloud-native platform that generates formatted actuarial XLSX reports on demand and on a schedule. Reports are defined by configuration rather than code: GEAR starts from a template workbook, runs the configured queries against the investment data lake, and writes each result set into its designated sheet with the template's formatting intact.
A FastAPI service in Azure Container Apps exposes HTTP endpoints through ingress and handles orchestration, while Azure Container App Jobs run the heavier generation work as containerized Python processes that produce downloadable XLSX reports. Separating lightweight API orchestration from compute lets reports be generated on demand across environments without long-running batch jobs or permanently allocated compute. The platform also provides Microsoft Entra authentication, secure browser and Teams access, release management through Azure DevOps, and event-driven notifications that alert users when monitored conditions are met.
The configuration-driven design is what let it scale — new report types are onboarded without redesigning the application. The automated report inventory nearly doubled over the past year to 69 production use cases, removing an estimated 95 person-days of manual effort per month and shifting that time from assembling reports to interpreting them. I won Talcott's ACE Award in 2026 for my work on GEAR.
GEAR — SLA Tracking & Operational Monitoring
Scheduled reporting depends on upstream files arriving on time. Daily and monthly delivery schedules, regional holidays, and different time zones made it difficult to distinguish a genuinely late file from one that was not yet due.
I built a configuration-driven SLA monitoring system for GEAR that checks file-processing logs against expected deadlines. The Python service supports daily and monthly business-day schedules, regional holiday calendars, time-zone-aware evaluation, and configurable grace periods. Each file is tracked by its originating SLA period, preserving outstanding obligations across day and month boundaries instead of overwriting them when a new period begins.
The system persists Pending, Met, and Violated states in SQL Server and records notification timestamps to control repeat alerts. Notifications are triggered only for confirmed violations, giving teams a consistent way to identify missed deliveries and investigate delays without manually checking file logs.
Financial Data Validation & Reconciliation Platform
Downstream reporting depended on data from multiple systems that frequently diverged due to timing differences, schema mismatches, and vendor inconsistencies.
I built Python validation pipelines that compare datasets across Oracle, SQL Server, and vendor feeds, normalize schemas, and surface discrepancies before they reach reporting workflows. The system runs automatically and produces actionable outputs, allowing teams to quickly identify and resolve data issues without manual investigation.
Metadata-Driven Ingestion Framework
Adding new data sources previously required writing custom ingestion logic for each format, slowing onboarding and increasing maintenance overhead.
I designed a metadata-driven framework where each data source is defined by configuration rather than code. The system handles XML, JSON, and Excel inputs using shared processing logic, enabling rapid onboarding of new sources while maintaining consistent validation, transformation, and error handling across pipelines.
Trade & Settlement Logging System
Trade and settlement activity needed to be captured, validated, and persisted in a way downstream processes could depend on — no gaps, no duplicates, with clear auditability.
I built the backend API routes and database logic to log trade activity, applying validation and upstream enrichment at ingestion time. I also added a React interface so the operations team could query and review settlement state without touching the database directly.
Optum 2020 – 2023
Enterprise Data Pipeline Development
Healthcare data arrived from multiple upstream sources in inconsistent formats, with no standardized validation layer before it reached analytics consumers.
I built and maintained ETL pipelines that ingested, transformed, and routed large healthcare datasets across internal systems. I focused on correctness at the transformation layer — enforcing business rules, surfacing upstream data quality issues early, and ensuring downstream analytics workflows received clean, reliable inputs.
Kafka Application Logging to Azure Event Hubs & ELK Stack
Application logs were scattered across services with no centralized way to monitor behavior, trace errors, or analyze patterns across systems.
I configured a Kafka-based logging pipeline that captured application events and streamed them to Azure Event Hubs. From there, logs were ingested into an ELK stack — Elasticsearch for indexing and storage, Logstash for transformation and enrichment, and Kibana for visualization. I also built Kibana dashboards to surface error rates, latency trends, and service health across environments.
Personal Projects
Trading Signal & Execution System
I built a Python system that monitors intraday market data and executes rule-based trades through Alpaca's brokerage API. It pulls a morning briefing from a Discord channel, builds a watchlist, and then monitors each ticker on a configurable candle interval — evaluating EMA crossovers and breakout conditions and submitting market orders with trailing-stop exits. PDT compliance is enforced automatically.
My goal was to understand the practical gap between signal generation and live execution: noisy data, timing constraints, and the difference between what a strategy looks like in backtesting and what it does under real market conditions.
Market Data Analysis & Research Tools
I wrote a set of personal scripts for analyzing historical stock performance, screening intraday setups, and evaluating valuation metrics across watchlists. They cover time-series pattern analysis, P/E screening, RSI-based momentum signals, and sentiment scoring from news feeds.