Avnish Kumar Email & Phone Number

Bangalore Urban, Karnataka, India

Bengaluru, Karnataka, India

Working on Content Recommendation Systems for Wynk Music and Xstream OTT--- Music Representation Learning using Multi Modal Contrastive Learning being used in a number of downstream applications.- A Music Topic Recommendation based on Acoustic Representation hosted on ES Hybrid HNSW Index. - An Acoustic Content Recommendation System to serve queries like "Hip-Hop Mix", "Flute Jazz Mix", "Sad Hip-Hop Mix".- Developed an Intent Sampling based First Song Prediction System.- Online re-ranking system for Wynk Music for First-Song-Prediction, Daily-Mix, and Other Collection reranking. - Experimented with point-wise, list-wise losses, and heuristic-based sampling to tailor the re-ranking algorithm for different use cases.Exposure to --- Methods - Word2Vec, ALS, xDeepFm, DLRM, DCN, Set-Transformers, VQ-VAEs, BERT, CLIP, Conformers, etc.- Technologies - Librosa, Pytorch, TF, MLFlow, PySpark, Axum, Airflow, Elastic Search, MongoDB, Redis.

Hyderabad Area, India

- Real-time S2T platform for Healthcare, Cricket, and Customer-Care.- Fine-tuned SOTA models like Wav2Vec2.0 for Hindi and English Language domain-specific datasets.- Scalable Threat Detection for Cyber Security using Kernel Density Estimation.- Telecom Supply Optimization using Quadratic Programming.- Territory Planning for Telecom Networks.- Robust Demand Forecasting for Telecom.- Geolocation Intelligence for mobility/migration tracing.

India

Worked on R&D for Data-Science Products.- Implemented DeepAR for Sales forecasting on AWS Sage-maker with Hyper-parameter tuning for robust time series learning.- Built a look-alike model for marketing using various multidimensional similarity matrices and designed evaluation metrics for the unsupervised task.- Implemented Random Cut Forest on Amazon Sage-maker for sales data streams on AWS.