Jaskaran Virdi Senior ML EngineerJaskaran Virdi Senior ML EngineerSenior ML engineer, eight years at Twitter, Rubrik and Imprivata. I get AI prototypes deployed, measured, monitored and cheap to run.
jaskaranvirdi1@gmail.com
What I can build for you
I can build the model, the platform under it, or both, on my own or inside your team. Each card shows where I have done the work before.
Agents that call your APIs, read your data and take real actions. Tool calling, guardrails, human review steps and a log of every step, measured with evals before and after launch.
Retrieval over your own documents with sensitive data stripped before indexing. OpenAI, Anthropic and Gemini behind one interface with automatic fallback, and streamed responses.
Eval sets and regression checks for ML and LLM systems, so you know whether a change made things better or worse. Monitoring by segment, tracing and dashboards once it is live.
Getting models out of notebooks and onto Kubernetes with Argo Workflows or Airflow. Job APIs, async serving, caching, memory tuning and smaller serving fleets, so the cloud bill goes down.
Classification and anomaly detection from problem framing to launch. Features, labels, evaluation, bias fixes and alerts that explain which behavior fired them. Including regulated settings like healthcare and identity.
Real-time voice with ElevenLabs text to speech and Deepgram speech to text, streamed so the reply starts speaking before the model finishes writing.
Systems I've built
Eight years at Twitter, Rubrik and Imprivata, plus an internship at Apple. Each in a domain where a wrong answer costs money or trust.
Built the job API and the orchestration workflows behind risk scoring, so other teams could launch scoring for each hospital customer, track it and collect results. Then ran the platform in production at scale and fixed the jobs that kept running out of memory.
A customer-facing assistant answers from company documents, with no personal data in the index.
Rebuilt from scratch. Customers got far fewer false alerts, and each alert shows why it fired.
Same detection, faster responses, and less load on storage.
Built and shipped models that detect toxic tweets across many languages and countries.
Fixed a training loop so the models learned from real reviews instead of their own output. Corrected a bias in the training data and added evaluation by user segment. Optimized a model to use a much smaller serving fleet.
Forecasting for iCloud and related web services.
Earlier: ML researcher at the Contextual Robotics Lab at UC San Diego, software engineer at Zomato.
Ways to work together
Remote, or on-site in San Francisco.
I go through your prototype, data, evals and infrastructure, then hand you a written plan to ship it.
I build and ship one system end to end. For instance an AI agent, a RAG feature or a model platform.
I join your team part-time or full-time and own a piece of the ML or AI roadmap.
Senior ML and AI engineering roles across LLM apps, agents, modeling and ML infrastructure.
Tools I use
The tools I build with, grouped by the kind of work.
Send me a few lines about what you have built so far and where it is stuck. Email works best. LinkedIn works too.
jaskaranvirdi1@gmail.com
Based in San Francisco, CA