Ayush Suhane Email & Phone Number

Bengaluru, Karnataka, India

Bengaluru, Karnataka, India

Delhi, India

Used generalized zero-shot extreme multi-label learning to build a product recommendation system. Increased performance especially on unseen labels. Reduced model latency from 350ms to 200ms by integrating a constrained decoder with zero-short ZestXML model. Built a api and streamlit app for the recommender for easy visualization and analysis of results.

Designed a Recovery System for containerized cluster architectures. The system design is based on MAPE-K feedback loop to build a self-adaptive system. The recovery system uses workload classification model along with environment simulation to detect the need for restart, vertical and horizontal scaling and take the recovery action. Designed a web UI for real-time monitoring of the cluster and recovery system. Orchestrated the recovery system using celery to make it asynchronus and scalable.

Bengaluru, Karnataka, India

Developed daily trading strategies using deep reinforcement learning. Trained RL models using RLlib and created trading environments using OpenAI Gym. Experimented with various deep reinforcement learning algorithms such as Advantage Actor-Critic, Deep Q-Net, DDPG, and Proximal Policy Optimization. Refined the trading strategies by using custom LSTM policy networks, reward shaping, and trading environment tuning. Deployed highly optimized and distributed hyper-parameter tuning on AWS cluster. Used Ax integration with Ray Tune to run hyperparameter trial.

Gurgaon, Haryana

Created a framework to identify and merge customer IDs that belong to the same customer. Used a graph structure for efficient querying and traversing of data. Identified and merged 4.5 million customer IDs.