• Designed and own the full-stack AI feedback loop: reviewers annotate model errors in a React/TypeScript UI I built; corrections flow through a Python/FastAPI pipeline into an eval system that restructures prompts and playbooks — improving recall from 55% to 92% after 15 annotated documents for a major enterprise customer

• Sole frontend engineer on the compliance product; own 100% of frontend code across the core product and multiple pilot applications built to close enterprise deals

• Reduced pipeline latency 3× (8 min → 3 min) by replacing LLM inference with Python post-processing where the model was being used for deterministic logic, then parallelizing the remaining independent stages

• Designed fault-tolerant Temporal workflows with idempotent retries, task queues, and parallel fan-out for correctness under bursty enterprise workloads

• Shipped a 0→1 production system on a 3-engineer team in 6 weeks; deployed full-stack pilot applications solo across government ACR contract review, pharma regulatory compliance, environmental marketing compliance, and pharma consumer review moderation

• Built a dry-run safety layer in Rails/PostgreSQL for high-risk administrative operations, enabling preview of database mutations before execution in production

• Built automated smoke-testing services to proactively detect production failures pre-release, reducing on-call incidents

• Improved CI/CD reliability with automated visual regression testing, reducing UI-related review cycles

• Designed multi-step LLM reasoning pipelines using LangChain for FinTech workflows, reducing financial brief generation time by 60×

• Developed dynamic system prompts refined iteratively with domain experts, improving output accuracy and consistency

• Built bulk review and evaluation tooling to support human-in-the-loop validation and accelerate prompt iteration cycles

• Built real-time computer vision pipelines (TensorFlow, Swift) integrated into a production iOS application for AI education

• Authored and deployed the majority of AIClub's computer vision curriculum, adopted by hundreds of students