Vaibhav Bhandari
Summary
Software engineer with 13+ years building production ML systems, developer platforms, and zero-to-one products. Spent a decade at Apple Maps — designed an ML platform adopted across the team, shipped aerial-imagery and NLP models to production, built an agentic AI platform, and led AWS infrastructure migration. Previously a founding engineer at a geospatial startup acquired by Apple. Currently prototyping AI-powered products using LLMs, embeddings, and agent architectures. Columbia OR background in optimization and stochastic systems.
Experience
- ML Platform & Agentic AI (2024–25): Designed and built a configuration-driven ML framework enabling data scientists to develop, deploy, A/B test, and evaluate models with minimal boilerplate — integrated with AWS EMR/Graviton, model reporting infrastructure, and PySpark. Adopted across the team for all new ML projects. Co-designed an internal agentic AI platform and ran training sessions enabling engineers to ship agentic use cases end-to-end.
- Aerial Imagery ML & Segmentation (2024–25): Built ML models from scratch for road-attribute detection using ortho imagery, probe data, and panoramic imagery. Created a generic segmentation pipeline enabling non-ML engineers to solve classification problems independently — first prototype classified terrain features from satellite imagery.
- ML-Driven Ops Automation (2022–24): Built classification models to predict no-action-required editorial tasks, reducing operations cost. Developed A/B framework giving Ops teams confidence to adopt ML automation. Targeted 10% total cost savings.
- Search & Data Infrastructure (2015–22): Designed a fast search and visualization service for multi-TB geospatial datasets (Spark + Solr + Kubernetes), creating world-build tile sets in ~1 hour. Built end-to-end address search and validation system. Led AWS migration — secrets management, S3 migration, Scala 2.12 upgrade. Owned production pipelines for map data processing across 20+ countries, driving 30%+ productivity improvements through automation.
- Mentored engineers across sites, led ML workshops and a Scala book club, and conducted hiring interviews throughout tenure.
- Designed and built a scalable, fault-tolerant search system for storing and querying multiple terabytes of geospatial data on AWS.
- Built an intelligent travel recommendation system using NLP techniques (LDA, LSI, keyword extraction) over multiple web sources.
- Worked directly with founders on product direction, architecture decisions, and go-to-market.
- Developed novel approaches for concept-drift prediction in energy load forecasting using hidden-state modeling and online ensemble learning with weighted majority algorithms.
- Built ML systems (Python) for GE's Ecomagination Challenge — electric vehicle load management using SVR, Random Forest, BART. Created a probabilistic model with Django/D3 visualization.
- Backend C/C++ development on an Order Management System. Built configurable firewall for algorithmic order execution and contributed to an ultra-low latency OMS prototype in Java.
Current Projects
Knowledge OS — AI-powered personalized Hacker News digest via WhatsApp. Sentence-transformer embeddings for semantic matching, author reputation scoring, engagement detection — backed by SQLite.
Memo Agent — Keyword-triggered knowledge base across WhatsApp, Telegram, and Discord with automatic metadata extraction and auto-tagging.
State Overflow — Newsletter applying statistical thinking (shrinkage, base rates, signal-to-noise) to everyday decisions.
Technical Skills
Education
Patents & Publications
Patents: "Total Property Optimization System for Energy Efficiency and Smart Buildings" (US 20150178865) · "Forecasting system using machine learning and ensemble methods" (US 14/707,809)
Publications: Di-BOSS (Fairmont Press, 2014) · Cost-optimal EV fleet charging via ML (IEEE SysCon, 2014) · Di-BOSS demo (NIPS 2013) · SVR for energy optimization (NYAS ML Symposium, 2012)