Mar 2026 — Present
Software Engineer Intern · Morningstar
- Own the production v2 rewrite of the team's AI rule builder, which turns analysts' plain-English data-quality rules into executable rule configs through a five-stage pipeline: the LLM only proposes structure, gated by verbatim evidence grounding, registry type-checks, deterministic compilation, and human resolution of ambiguous datapoints.
- Built its evaluation harness, grading LLM output by semantic equivalence to hand-parsed ground truth and requiring refusal on underspecified prompts, with per-call cost accounting and SSM-backed secrets handling.
- Architected auto-scaling AWS infrastructure (CDK, Fargate, ALB) for a full-stack RAG chatbot, with DynamoDB powering real-time user sessions and low-latency LLM context retrieval.
- Ship UI features and REST APIs for Illume 2.0, a modular RAG document-extraction platform serving 2,000+ non-technical users across 5 global offices.
- Cut data-collection turnaround from a ~5-week manual JSON build-and-test cycle to under a minute with UI-driven pipeline configuration, across 1,000+ financial documents per workflow.
- Co-developed a proposed investment-quality metric and built its prototype — presented to the CEO, CTO, and senior leadership, now in company-wide research for implementation.
- Engineered the core logic of a regression-testing suite (CDK + SQS) that batch-queues pipeline runs and computes recall and F1 across variants to drive improvements.
- Built and debugged Harness CI/CD pipelines deploying to Kubernetes for staging and production releases, and a New Relic performance dashboard for Illume Studio tracking p50/p95 page load, API latency and failure rate, and JS errors.
- Retained for a fall-term extension, shipping platform features alongside a full course load.
PythonTypeScriptAWS CDKDynamoDBECS FargateKubernetesRAG · LLMNew Relic