Ardent Privacy, enterprise data discovery
An LLM gateway in Go that holds 5,000 req/s
M.S. Computer Science, UC Santa Cruz
faster API responses on the Data Discovery backend, from query tuning and Redis caching
p95 overhead added by my LLM gateway at 5,000 requests a second
recall@5 on a RAG service over 7,600 documents, versus vector-only search
to fail over to another model provider during simulated outages
Backend for the flagship product that lets enterprises scan, classify and manage sensitive data, with compliance logic and the APIs the frontend runs on.
Apps that call model APIs directly inherit provider outages, runaway costs and leaked sensitive data. This gateway sits between services and providers and absorbs all three.
Keyword search misses context and LLM answers go unverified. This API answers over 7,600 documents and scores its own quality.
A distributed incident response platform that orchestrates several AI agents through a checkpointed workflow that can resume where it stopped.
Routing, rate limits, caching, redaction, failover. These are the parts nobody demos, and the parts that decide whether a product stays up. My gateway handles them in one place so each service does not have to.
I spent two years at Ardent Privacy building the backend for a data discovery product, where sensitive data, compliance rules and large record counts leave little room for slow or fragile services.
Outside that job I build the infrastructure around language models: a gateway that survives provider outages, a retrieval service that measures its own answer quality, and an agent workflow with approval gates.
I’m now studying for an M.S. in Computer Science at UC Santa Cruz, after a B.E. in Computer Science from Savitribai Phule Pune University (3.81/4.00).
Backend and APIs for the Data Discovery product, plus CI/CD and deployments.
Built a mock Gmail API in Rust with the Gmail API and OAuth, and learned Docker deployment.
Flutter and Firebase app that explains dashboard warning lights. Cut issue identification from 55 to 25 seconds.
Algorithms, advanced data structures, network security, machine learning, NLP, data visualization.
Deep learning for posture monitoring and yoga pose refinement. Presented as “PoseCraft Fitness Navigator” at ICRTACT-2024, an international conference.