Experience
Industry, research, and teaching roles across search & recommendation, real-time computer vision, edge ML, and backend engineering.
-
AI Engineer Intern, Search & Recommendation Systems | Tabhi- Architected an end-to-end personalized discovery feed for 450K+ items using hybrid retrieval and slot-based re-ranking.
- Designed recency-decayed personalization over Elasticsearch BM25 and pgvector retrieval using behavioral signals.
- Reduced cold-feed latency from 8 s to 500 ms by three-tier caching, cache pre-warming, optimized indexing, and geospatial retrieval.
- Built recommendation analytics using RudderStack to capture user behavior and optimize personalization through engagement metrics.
- Resolved Elasticsearch–MongoDB consistency issues and hardened location-aware retrieval, improving recommendation reliability.
ElasticsearchpgvectorHybrid RetrievalRe-rankingCachingRudderStackMongoDB -
Research Assistant, Real-Time AI Systems | University of Southern California- Implemented a monocular SLAM pipeline for assistive smart glasses using a 16-zone ToF sensor to recover metric scale.
- Built a GTSAM pose-graph backend in C++, reducing trajectory error from 29–58 m to 0.09–0.24 m while optimizing 2,486 poses in 0.36 s.
- Designed a physics-based ToF simulator and optimized zone pooling, halving trajectory error with 2–4% scale accuracy.
- Extending the factor graph to fuse IMU, ultrasonic, and GPS measurements for deployment on a Jetson Orin Nano.
SLAMGTSAMC++ToF SensingSensor FusionJetson Orin Nano -
Grader, CSCI 585 Database Systems | University of Southern California
- Evaluate Database Systems coursework for 200+ students across core and distributed systems.
- Develop and grade technical assessments on database algorithms and architectures, evaluating correctness and performance tradeoffs.
Database SystemsDistributed SystemsTeaching -
Machine Learning Intern, Edge Computer Vision | YMT Medical
- Redesigned YOLOv8 with a P2 head and NWD loss for micro-lesion detection across 2,000+ patient images, achieving 0.83 mAP@0.5.
- Built a Laplacian/HSV image-quality pipeline to detect blur and glare in patient uploads before model inference.
- Optimized edge inference using INT8 quantization and operator fusion, reducing latency by approximately 40%.
- Mapped 468 MediaPipe facial landmarks to compute deterministic GAGS acne scores offline with an active learning loop.
YOLOv8INT8 QuantizationMediaPipeOpenCVEdge ML -
Software Engineer Intern, Backend & APIs | Technoriya ERP Solution- Developed Python and Node.js backend services and APIs supporting scalable data pipelines and cloud-based ML workflows.
- Implemented secure authentication and access protocols via Firebase Auth and Firestore rules to ensure data integrity.
- Optimized RESTful APIs and Firestore queries to achieve 450 ms response times under 1,000 concurrent requests for ML integration.
PythonNode.jsRESTFirebase
Awards & Recognition
- 1st Place, Syrus 2023 Hackathon
- Second round, Smart India Hackathon (SIH) 2023
- Finalist, IndeHub Hackathon 2025
- CodeShastra 9.0 & X, DJ Sanghvi College of Engineering
- Presented at IEEE ICSCSS 2023 · SWE Conference 2025