Where I've worked

Built a simulation platform to let product teams test copy, features, and pricing against AI-generated customer personas — shipping decisions in minutes instead of weeks.

• Shipped MVP in 3 weeks on FastAPI, Postgres (pgvector), Redis, and Docker with CI/CD
• Built an LLM-driven persona engine with clustering that turns qualitative reasons into actionable recommendations
• Ran pilots and back-tested against historical A/B tests showing 60–70% predictive accuracy
• Led GTM across ICP definition, pricing tests, waitlist-to-beta funnel, and design-partner onboarding

Built an adaptive assessment co-pilot that asks one question at a time, analyzes the answer, and targets each student's mastery gaps in real time.

• Piloted at local tuition centres, demonstrating measurable increase in learning efficiency
• Conducted field interviews with educators — uncovering tech-averse faculty, unreliable infrastructure, and crowded LMS market dynamics

Built a peer-to-peer rental platform for high-value gear, validated through interviews and pilots across India's scuba diving communities.

• Facilitated smooth transfer of gear across India's scuba diving communities
• Validated product-market fit through customer interviews and live pilots

Led two national-scale ML deployments: a real-time disease outbreak detection pipeline for India's NCDC and a pest management system serving thousands of farmers.

• NCDC Outbreak Alert Pipeline: deployed a multi-stage, multi-lingual detection pipeline (rule-based filters → BERT → LLMs → clustering) scanning 400M+ media sources, issuing 100K+ alerts, and verifying 5K+ outbreaks. Published at ACL 2025
• Pest Management: trained ResNets, InceptionNets, and EfficientNets for pest detection on cotton and rice farms; advanced augmentation improved robustness by 14%, directly helping 10,000+ farmers

Built and deployed ML systems for TB treatment adherence prediction and oral cancer screening at national scale.

• National TB Adherence Prediction: built a similarity-encoded ensemble (XGBoost, LightGBM, EBM) on 0.8M Nikshay TB cases — AUC ROC 0.80, Recall@20 of 0.62 (3× baseline). Deployed across 15 states, flagging 100K+ high-risk patients. Best Paper at ML4H @ NeurIPS 2022
• Diffusion-Augmented Oral Cancer Screening: developed a lightweight CNN for early oral cancer risk; diffusion-augmented training lifted recall from 79% to 91%. Co-led data ops with PGIMER Chandigarh on 10K+ labelled oral lesion images

Built large-scale record-linkage systems for India's National Digital Health Mission and the CoWIN vaccination platform.

• Probabilistic Health ID Matching: de-duplication engine using TF-IDF blocking, MinHash-LSH, and GBDT — merged 90M+ patient records with 96% precision and 92% recall, cutting manual review by ~80%
• Re-tooled the pipeline for CoWIN to screen sign-ups and block duplicate vaccination attempts during India's nationwide COVID vaccination drives