Jul 2025 – Sep 2025 · 2 mos · Greater Bengaluru Area · On-site

• Ads trust

Jul 2024 – Jul 2025 · 1 yr · Hyderabad · Hybrid

• Azure OpenAI Training & Finetuning Service

Feb 2023 – Jul 2024 · 1 yr 5 mos · India · On-site

• Image Ranking, Hotel Ranking & Generative AI (Deep Q Learning, Bandits, Counterfactual Learning to Rank, Multimodal generative ai)

Sep 2021 – Feb 2023 · 1 yr 5 mos

• Video Attribution Removal: logo detection, semantic segmentation, video-inpainting, opencv Morphological operations
• Video Commerce: Modified yolo v5 architecture and loss function to beat SOTA architecuture (Multi-task Mask RCNN) for Fashionpedia Dataset, MultiModal search, Self-supervised Learning (SimCLR), Object Recognition (Arc Face Loss), FAISS (embedding search and retreival), Detectron, Mixed Precision Training, Pytorch Lightning
• Image/Video SuperResolution: No-Reference Image Quality Assessment: Paq2Piq finetuning (Collaboration with Prof Bovik, UT Austin), Unsharp mask, anti-aliasing, Real ESRGAN (quantised), Microservices Architecture (Pub/Sub, Deadlettering, Multi-processing), Performance Monitoring: Grafana, Kibana

Mar 2020 – Sep 2021 · 1 yr 6 mos · Gurugram, Haryana, India

• Computer Vision: Semantic Segmentation and Opencv preprocessing based pipeline, Facial landMark Localisation (2d/3d) & Tracking, Eye blink detection & gaze tracking, 3D Head Pose Estimation, Mediapipe & VR
• Deep Learning: Uncertainty Aware MultiTask Learning (Single-Shot HikeEmoji Generation)
• Model Optimisation: Knowledge Distillation, Quantisation, TFLite
• Recommender Systems: Contextual MultiArmed Bandits
• Developing ML model pipeline and serving it at scale.

Jun 2018 – Mar 2020 · 1 yr 9 mos · Gurugram, Haryana, India

• Similar and Upsell Product Recommender Rail
• Key terms: Image based Similarity, Contrastive Learning, Empirical Bayes, LightGBM, Duplicates Removal
• RTO - predicting Return probability of the user-product pair given the static, dynamic and rolling features of user, product
• Key terms: Feature Engineering, Data Engineering, LightGBM with 200+ features, character based BiLSTM model, Attention based model, BERT, Distil BERT
• RecSys, feed ranking- LightFM, Neural Network based recommendation
• Data Engineering- Pyspark, EMR, Apache Hadoop
• Backend Caching Service for ML pipeline (MultiThreading java service for caching parquet files to Aws s3), Deployment of ML models at Scale

Jun 2017 – Jul 2017 · 1 mo

• Built claims incidence model, severity model and fine tuned fraud detection model using Random Forest , gradient boosting and Xgboosting classifiers. Later expanded on Azure machine learning studio.