Staff Machine Learning Engineer, Perception, Autonomy at Rivian

Led the backend and ML delivery of an in-vehicle conversational assistant, using Python (FastAPI) services to turn voice intent into safe, traceable actions across vehicle controls and diagnostics, with OpenTelemetry traces and CloudWatch alarms baked in from day one. Built the agent/tool orchestration layer with FastAPI + gRPC, routing requests to internal services and enforcing hard safety gates, while exposing partner-facing endpoints through REST + GraphQL with strict contracts. Integrated Gemini on Google Vertex AI for higher-reasoning turns, but kept action execution in FastAPI/Node.js services so the LLM never 'freestyles' side effects — everything goes through audited tool calls. Implemented grounded responses with LangChain + Pinecone so the assistant pulls from trusted knowledge and live state before answering; persisted long-lived context in PostgreSQL and hot session state + rate limits in Redis. Built a telemetry ingestion pipeline on Kafka (MSK) that streams events into

Sr. Machine Learning Engineer at Meta

Built DLRM-style ads ranking models in PyTorch using big sparse embedding tables plus dense features to predict CTR/CVR and drive final ad ordering. Trained and iterated the models with distributed PyTorch (DDP) and mixed precision, and kept runs stable with solid checkpointing, reproducible configs, and clean data sharding. Improved ranking quality by tuning feature sets and embedding strategies in PyTorch and validating every change with offline evaluation plus online A/B testing. Worked on candidate and training data quality by tightening negative sampling, cleaning labels, and adding leakage checks so PyTorch training wasn’t learning 'shortcuts' from logging artifacts. Kept production inference fast by optimizing PyTorch serving, tightening request batching, reducing expensive feature retrieval, and using caching for hot lookups when it actually helped latency. Set up regression coverage for launches by wiring offline eval gates to the release process and tracking model quality/l

Software Engineer – Machine Learning at Cresta

Built the “Expertise Engine” real-time Agent Assist product with Python (FastAPI) and Node.js, delivering live coaching and next-best suggestions during voice/chat. Trained and shipped NLP models in PyTorch (transformers like BERT/GPT-2/T5) for intent, objection, and moment detection to trigger the right coaching at the right time. Used gRPC to keep the real-time inference → routing → UI path consistent, even with noisy/partial transcripts. Built voice modules with custom ASR and entity recognition, then aligned context so suggestions stayed grounded in what the customer actually said. Exposed orchestration through REST + GraphQL with OAuth2 and RBAC, and stored durable session state in PostgreSQL with Redis for low-latency context and rate limits. Streamed events through Kafka (MSK) and stored logs/training datasets in S3 to support retraining and offline evaluation. Deployed with Docker on Kubernetes (EKS) via Terraform and GitHub Actions, and monitored production with OpenTe

Machine Learning Engineer at Woebot Health

Designed and implemented highly accurate NLP classifiers with state-of-the-art deep learning techniques. Led and implemented strategies for obtaining high-quality training datasets. Defined and analyzed product metrics and built dashboards to visualize company health and enterprise campaign performance. Provided mentorship and guidance to team members on machine learning projects.

Data Scientist at Branch

Part of a fast-paced and agile team that iterated through multiple versions of proof of concepts and delivered to potential partners. Built complex search and recommendation systems and touched components across the stack, including API Services, Backend Data Storage, Search Querying Logic, and Frontend Web Applications. Conducted experiments with A/B testing to measure engagement lift over other competing products.

Software Engineer at Branch

User feature modeling and engagement prediction with machine learning. Led analytics backend for partner facing dashboard products. Developed analytics system for handling a high throughput, real-time data stream. Mined business insights from large data sets for data driven decisions. Product re-engagement and up-sell to customers with machine learning.

Summer Fellow at Lightspeed Venture Partners

Full stack developer for a web application that significantly improves the trip planning experience. Also developed the iOS version of the same app.

Design Engineer at AMD

Developed software infrastructure for optimizing simulation performance for CPU core verification.

Senior Machine Learning Engineer at Meta

Built DLRM-style ads ranking models in PyTorch with large sparse embeddings and dense features to optimize CTR/CVR and drive final ad ordering. Trained with distributed PyTorch (DDP) and mixed precision; maintained reproducibility with proper checkpointing and sharding. Improved ranking through feature engineering and offline eval gates; validated changes with online A/B testing. Ensured fast inference by optimizing serving, batching, and caching; established regression coverage by linking offline eval to releases and monitoring model quality/latency in production. Collaborated with ads delivery and platform teams to land changes safely in the auction environment; built data guardrails for training/serving consistency; mentored engineers on embeddings and evaluation.