Jun 2022 – Sep 2025 · 3 yrs 3 mos · Bengaluru, Karnataka, India

• Promotions -
• ML Engineer 1 -> ML Engineer 2 -> Senior ML Engineer
• 1. Semantic Search and Ranking:
• Developed a Cross encoder framework with PyTorch and BERT-based architecture for encoding product information and search queries.
• Implemented a proof of concept using Typesense Database for search candidate generation and improved Autosuggestion corpus for better click depth ratio.
• Tools: PyTorch, NLTK, Typesense, FastAPI, Docker.
• 2. ETA Optimization:
• Enhanced delivery time accuracy by 120 seconds and retention by 12% using a retrained XGBoost model.
• Established an in-house feature store with Kafka, Spark Streaming, and Redis for real-time store feature generation.
• Built a retraining pipeline for Iceberg data lake integration and mentored an intern to develop an in-house ETA maps prediction algorithm.
• Tools: Python, Scala, FastAPI, Spark, Iceberg, Kafka, AWS, Docker, Kubernetes.
• 3. Supply Chain Forecasting & Optimization:
• Implemented efficient data and ML pipelines with EMR, Glue, and PySpark for inventory time series forecasting.
• Developed a vendor-scheduling service for optimized resource allocation and order purchase.
• Tools: Python, PySpark, FastAPI, Airflow, Redshift, AWS.
• Skills: Python, SQL, Pyspark, Spark streaming, Kafka, Redis, Delivery time Prediction (real-time service & ML Platform), Time Series, Supply chain optimization, Fraud Modelling

Feb 2022 – Jun 2022 · 4 mos · Bengaluru, Karnataka, India

• Extracting security details from Redash(dig into the internals of Redash's codebase that was hosted in the network).
• Analysed User SMS data for business needs.

Jun 2021 – Aug 2021 · 2 mos

• Integrated the Jax framework into Deepchem codebase which interacts with the existing modules suited for made for bio-informatics and analytics.
• Revamp the CI/CD pipeline for the organisation into three separate setups (also followed by libraries like Huggingface, Atari, etc ) due to the dependency inconsistencies with Tensorflow and other libraries, etc.
• Built a general framework for solving differential equations using Neural Networks with the help of the Jax framework built during the same period inspired by Physics Informed Neural Networks.

Jun 2021 – Aug 2021 · 2 mos · Tokyo, Tokyo, Japan

• Built an end-to-end face recognition system used for attendance monitoring connected that could be connected to Vido Surveillance Camera or WebCam. Deployed the full pipeline in AWS server using Flask and also provided a RestAPI for local prototyping. Connected the final application to Tableau for further visualization.
• In the Later half, we expanded the system to include Anamoly Detection with the model using unsupervised training (Multi-instance Learning).

Nov 2019 – Nov 2020 · 1 yr · Roorkee, Uttarakhand, India

• Built an ensemble ML model to predict potential inhibitors of the SARS coronavirus protease molecule. The ensemble was built on models trained using Graph Neural Networks, Neural Networks, Tree and Boosting methods. Was able to achieve an average AUC-score - 0.76 and an average PRC-AUC - 0.301 on 5 folds. Were able to predict 12 potential drugs after docking.
• Applying deep learning techniques on pharmacore fingerprints for Alzheimer's disease with Biomolecules from PubChem dataset and drugs from the Drugbank dataset. Had to work with a highly imbalanced dataset and applied focal loss function. Obtained a 0.89 F1 score on Test data and predicted 107 potential drugs for further testing.