Hi, 

Hybrid RAG + Agentic Research Platform

Built an enterprise AI research platform (think Perplexity, but for science) — used daily by lab teams to search the scientific literature, look up drug compounds, and explore hypotheses. Became the primary AI interface across the organization within weeks of launch.

Designed a hybrid search system that combines traditional keyword matching with modern semantic search — so it finds the right documents whether you phrase a query exactly or ask it conversationally. Achieves high recall across large chemical and biological literature corpora. (Technical: BM25 + dense embeddings, late fusion, cross-encoder reranking.)

Built a long-context memory system that lets the AI maintain coherence across multi-session research workflows — so scientists can return hours or days later and pick up exactly where they left off, with the AI retaining full context. Scales to millions of words without losing coherence.

Gave the AI the ability to autonomously use specialized tools — toxicity lookups, molecular property prediction, chemical structure searches, and live queries to scientific databases (ChEMBL, PubChem) — so it can answer complex drug discovery questions without a human in the loop.

Deployed a 27 billion parameter language model on our own AWS infrastructure — no reliance on external AI providers, full data control. The system is modular, so newer and more capable models can be swapped in as they become available without rebuilding the application.

PAL-3 Post-Hoc Clinical Analytics · VistaGen Therapeutics

Investigated why a large Phase 3 clinical trial for social anxiety didn't show the expected improvement over placebo. Combined patient health scores over time, site visit records, enrollment timing, and speech patterns measured from recorded therapy sessions to find out what actually happened.

Discovered that how patients spoke during visits — how much they talked, sentence length, total talk time — was a reliable predictor of who responded to placebo. This pointed to a habituation effect: the act of repeatedly attending clinic visits was itself reducing anxiety, making it harder for the drug to show a measurable difference.

Also explored how recruitment channels, site scheduling patterns, and patient vs. clinician communication styles influenced outcomes — delivering concrete recommendations to help the sponsor design more robust future trials.

Financial Risk & Pricing ML · DriveTime · Carvana

Led development of production ML systems for financial risk and pricing products used by DriveTime and Carvana — two of the largest used-car finance platforms in the US. These models directly influenced which loans get approved and at what rate, across hundreds of thousands of originations annually.

Built models that predict loan pricing, credit risk, vehicle value over time, and likelihood of default — achieving F1 scores close to 0.90 across use cases, validated through rigorous backtesting against historical loan performance.

Owned the full data pipeline — working closely with client teams on data quality, feature selection, and model validation. Also ran fairness and bias audits to ensure the models didn't discriminate in ways that would create regulatory exposure.

Data Engineering · Warehousing & Lake Pipelines

Built and maintained batch data pipelines and analytics warehouses that feed underwriting and pricing models — landing raw data in Amazon S3, orchestrating transforms with AWS Glue and PySpark, and delivering curated features and reporting layers in Amazon Redshift for analysts and model consumers.

Owned schema evolution, partition strategies, job monitoring, and cost-aware ETL so daily scoring and backtests stayed reliable as data volume grew across client fleets.

ML Inference Platform · Kubernetes · IaC

Led the team to build a scalable microservices inference platform on AWS Kubernetes — packaged with Helm for repeatable releases and provisioned with Terraform (infrastructure as code). Full observability via the ELK stack — structured logging, latency dashboards, and auto-scaling to absorb burst inference across multiple model families.

Authored custom Docker images with CUDA support and GPU-aware runtimes so deep learning models serve with predictable performance in production.

LMS Rebuild · Full-Stack Platform [Case Study]

75% improvement in Core Web Vitals — led the rewrite of a large-scale Learning Management System serving 3M+ active learners, with a strong focus on PWA support, offline-first usage, and global responsiveness for low-bandwidth networks.

Architected the stack across React + TypeScript for learner UX, Node.js and Golang services for performance-critical APIs, and caching layers with Redis. API response times were pushed below 200ms for core learner journeys.

Implemented edge-aware delivery and multi-geo cache strategy (regional CDN + Redis invalidation patterns), so content stayed fast and consistent across markets. Combined code splitting, prefetching, and fallback offline flows so learners could continue progress even with unstable connectivity.

Worked closely with product, design, and infrastructure teams on rollout, SLO tracking, incident response, and platform guardrails; mentored engineers through the migration and established performance standards that persisted after launch.

Assessments & Test Platform

Revamped the online assessment and test platform used for proctored exams and skills validation at scale. Rebuilt critical paths for faster item load, more stable delivery under spike traffic, and clearer observability for the ops team — with up to ~40% performance gains on slow-device profiles in lab tests.

Same React / TypeScript / Node stack as the LMS program, extended to secure session handling, timer-critical UX, and coordination with backend teams on data consistency for grading pipelines.

ML Product: “Shorts” — Micro-Learning Feed

Engineered upGrad Shorts as a learning-science driven micro-learning product. Built the ranking and retention engine using the SM2 spaced repetition algorithm paired with a neural network classifier to predict optimal review intervals per learner, improving long-term knowledge retention.

Owned the end-to-end experience across personalization logic, experimentation loops, and React delivery surfaces for short-form lessons, balancing educational outcomes with feed engagement.

Partnered with Growth on acquisition, retargeting, and cross-sell — the feature drove a measurable lift in engagement and contributed roughly 15% improvement across retargeting and cross-sales experiments tied to the short-form experience.

Yahoo Ad.com — Self-Serve SMB Ad Platform

Contributed to Yahoo Ad.com, a self-serve advertising platform enabling SMBs to create and manage ad campaigns across the Yahoo network with minimal setup friction.

Built smart onboarding flows — a small business just enters their website URL, name, and category, and the system automatically generates a ready-to-launch campaign. Measurably reduced the time for a new advertiser to get their first ad live.