Work
Five companies, five very different problems. In every one of them, the modeling turned out to be the easy part. Understanding what the problem actually was, and keeping the thing alive in production, was the work.
Meta
Meta's internal operations teams run on data at a scale most companies never see. I work on AI systems that help leadership make faster decisions: NLP pipelines that turn meeting transcripts into structured insights, a natural-language-to-SQL agent for real-time data retrieval, and operations research models that optimize how resources get scheduled across facilities. I also build simulation models that stress-test supply chain and staffing scenarios before high-traffic events.
CXAI
CXAI builds enterprise workplace software for Fortune 500 companies. I owned the AI layer: RAG pipelines, LLM fine-tuning, vector search, ML forecasting, and the backend infrastructure to serve it all to 100k+ weekly active users. Fine-tuned LLaMA-2 with QLoRA on 2M+ support logs, and cut inference costs 25% by optimizing GPU batch serving with vLLM. Built on Qdrant, FastAPI, Postgres, and GKE. The latency target was sub-200ms. We hit it. Users never noticed, which was the point.
TruckX
TruckX builds telematics software for commercial trucking fleets. I worked on ML systems for battery health monitoring and a recommendation engine that figured out where drivers should stop. The battery degradation algorithm came from first principles: analysing degradation curves across lithium-ion and solar-panel chemistries, then setting thresholds that cut unexpected failures by 32%. The stop recommendation system went to a 300-truck pilot and cut fuel usage by 15%.
Dream11
Dream11 is India's largest fantasy sports platform, with a 200M+ user base and peak traffic during IPL. I built optimization models using Gurobi for contest structuring and team selection, churn prediction with XGBoost, and causal inference work on promotions. A/B tests I ran on pricing and bonus structures increased revenue per user by 10%. Fantasy sports turned out to be full of optimization problems wearing other names.
OYO
OYO was the world's third-largest hotel chain when I joined. I built the analytics infrastructure: automated pipelines, real-time pricing algorithms, and Python bots that delivered CXO-level metrics to Slack every morning. Reduced reporting turnaround by 90%. Learned early that the difference between a useful model and a useless one is whether someone actually looks at the output.
MS Data Science, University of San Francisco
Focused on the applied side: A/B testing, causal inference, distributed systems with Spark, and statistical modeling. USF's program is hands-on by design. Most of the work was real datasets with messy, real problems.
BE Information Technology, GGSIPU University
Where I learned to build things from scratch before frameworks did everything for you.