Prag Batra

- Subscription growth and monetization team at X which was acquired by xAI in March 2025 - Used Grok to triple the number of premium business signups and corresponding subscription revenue through targeted upsells - Designed, implemented, deployed and maintained server-driven framework for unified upsell serving and tracking for millions of users across platforms including offer targeting, swapping and rotation, dynamic metrics display and filtering criteria, and ad-free trials - Designed, implemented, deployed and maintained upsells, immediate and delayed reminder notifications and backend across millions of users for dynamically and uniquely assigning offer codes at scale for a NordVPN trial included with eligible premium subscription signups - Added support for 3 and 6 month premium basic, premium and premium+ subscription gifting

- Fine-tuned and deployed improvements to in-house LLM to automate manual back-office processing of PDF medical record request letters - Tripled % of request letters matched to the correct requester entity from 20% to 60%, matching an additional 16,000 request letters per day - Added field logging for 3 new types of request letters, tripling the total number of fully logged request letters (additional 14.8k letters fully logged per day; increase from 15 to 32% and 55 to 75%) and almost doubled the number of partially logged request letters from 15k to 28k - Integrated Google Maps API for address validation and implemented hallucination detection on predicted fields such as patient name and fax number - Reduced overall cost by ~$3,200/day (<$0.03 per fully logged letter vs ~$0.25 per partially logged letter with human completion)

- Personalized AI travel itineraries & local experiences iPhone app with 8,000+ users - Worked closely with CEO to identify and develop product vision through multiple pivots, MVPs, UI/UX reviews, and beta tests with users - Applied to YC S13 and our application was identified as being in the top 10% - Experimented with AgentGPT and AutoGPT to add a personalized AI chat to the app which could generate more detailed travel itineraries accounting for current flight and hotel prices/availability, etc. - Designed and built Firebase-hosted Bootstrap HTML/JS itinerary sharing page connected to Firebase NoSQL DB and Python web scraper to extract key events and experiences from a given webpage using GPT3.5-16k for use in personalized local experience recommendations

- Designed and implemented key product features based on customer feedback including printing summary data such as institution name on each output image frame, supporting additional DICOM image formats (DICOM-GSPS, SC, and SR), and allowing customizability of output data and watermarks - Created SLOs and SLIs, observable metrics (via Prometheus and Grafana), and alerts for key events in production pipeline - Designed and implemented multi-level anonymizer exceeding HIPAA requirements to protect patient data while maintaining inference accuracy - Increased dev velocity by an order of magnitude + Designed & built internal libraries & frameworks for common tasks including serialization to upload & download data from S3 + Generated and sent customizable multi-format DICOM output files + Led multiple large-scale refactoring tasks across our entire codebase including unit, integration & end to end tests on CircleCI - Led multiple efforts to streamline real-time distributed production pipeline involving 1-2 petabytes of data + Reduced real-time inference turnaround time from 10-15 mins to 10-15 s by using GPU nodes on EKS + Reduced S3 upload and download times from minutes to seconds by optimizing output datatypes + Reduced frontend load times from minutes to seconds and enabled multi-user access by resizing, chunking, and compressing image slice data prior to URL pre-signing - Interviewed, hired, and onboarded new engineers as needed to build out the team - Wrote portions of verification and validation (V&V) test documentation for FDA software as a medical device (SaMD) application - Involved in design, development and deployment to production of PyTorch R-CNN machine learning model based on Inception v2 to improve inference accuracy of another highly commercializable model

Independent development of a variety of iPhone apps ranging from an interactive homework viewer to a resolution voting platform to a location-based call forwarding utility for Google Voice.

- SAP HANA Innovation Award for Medical Record of the Future, 1st place, Next-Gen Apps (May 2016) - Developed Sequence to Medical Phenotypes (STMP [https://github.com/AshleyLab/stmp]) bioinformatics pipeline for analyzing genome sequence data for individuals with rare, undiagnosed diseases. Features included variant annotation, prioritization, pharmacogenomics, and tools for analyzing genomic trios (mother, father, child). - Pipeline was primarily Python with Cython HMM for trio analysis, Bash scripts and third-party bioinformatics tools (eg annovar, bedtools, snpEFF), YAML configuration files, and SQL database support (MySQL and SQLite). Briefly worked with SAP HANA db technology. - Wrote generic Bash wrapper scripts for both SGE and SLURM job submissions to provide common interface to run STMP and other jobs

Designed and prototyped a medical device to assist with spinal surgery as part of a collaborative effort between Stryker Corp. and the WashU School of Medicine Center for Innovation in Neuroscience and Technology (CINT). Presented to engineers and executives at Stryker.