Looking for a job

vikas kumar

@vikaskumar23

Message

Data scientist at Citi building a RAG chatbot that uses semantic search to improve LLM knowledge retrieval.

India

Message

At Citicorp Services India Private Limited, I built a Sales Assist chatbot with a Retrieval-Augmented Generation pipeline. I also automated data ingestion into Amazon S3 and used embeddings and semantic search to improve knowledge retrieval.

At Pocketfm, I worked on feed personalization and show ranking. My recommendation strategies increased user engagement by 4%, while cold-start recommendations based on projected revenue per listener delivered a 10% uplift in average revenue per user.

At Gojek, I implemented a central store for reusable offline features and designed A/B experiments for text search ads. One experiment improved ad revenue per session by 20% and was statistically significant.

At Infrrd.ai, I developed machine learning approaches for document information extraction, classification, and object detection, and owned an ML module from proof of concept through commercialization. Earlier, I worked on solar concentrator and receiver optimization at Indian Institute of Science and Masdar Institute of Science and Technology.

Experience

Work history, roles, and key accomplishments

Current

AVP - Data Science

Current

Citicorp Services India Private Limited, CitiGroup

Jan 2026 - Present (9 months)

Built and automated data ingestion pipelines to integrate external data sources into Amazon S3, enabling scalable data processing and analytics workflows. Developed a vector-based Retrieval-Augmented Generation (RAG) pipeline using embeddings and semantic search to enhance LLM response accuracy and knowledge retrieval.

RAG S3 Data Pipelines Embeddings Semantic Search LLM

Staff Data Scientist

AngelOne

Mar 2025 - Dec 2025 (9 months)

Architected predictive models for First-Year Revenue with 6% MAPE across large user base, enabling precise customer segmentation. Led ML model deployment and scaling practices to serve business-critical forecasts in production systems.

Machine Learning predictive modeling Model Deployment Scaling MAPE

Senior Data Scientist

Pocketfm

Jan 2023 - Mar 2025 (2 years 2 months)

Designed and executed strategies to optimize content recommendations, leading to a 4% increase in user engagement. Used auto-regressive data structure to predict ARPL@hN+10 based on ARPL@hN and achieved MAPE of 2% across show language, locale, gender.

Recommendation Systems Personalization ARPL Prediction Autoregressive Models MAPE

Data Scientist

Gojek

Jul 2022 - Nov 2022 (4 months)

Implemented a central store to promote re-usability of offline features across multiple use cases and act as SSOT for offline model training and online feature serving. Proposed and designed ML architecture to introduce optimisation of bookings in text search ads ranking, and designed and analysed multiple A/B experiments.

GitLab CI CD BigQuery Airflow A B Testing Feature Stores ML Architecture

Data Scientist

Infrrd.ai

Jul 2018 - Jun 2022 (3 years 11 months)

Worked on intelligent Data Capture team, created ML architecture for training and testing loop for document information extraction from scratch to augment existing rule-based system. Owned End to End ML module from proof of concept to commercialization.

Machine Learning Deep Learning CNN Object Detection Document Classification Data Annotation

Data Scientist

Tata Group Technology Innovation Office

May 2017 - Feb 2018 (9 months)

Identified correct data required for solving business problems, cleaned and pre-processed datasets for predictive models, and performed exploratory data analysis. Applied machine learning techniques to process data for Flisom's thin film manufacturing and boiler tube leak detection.

Machine Learning PCA EWMA Data Cleaning EDA predictive modeling

Research Assistant

Indian Institute of Science

Sep 2015 - Feb 2017 (1 year 5 months)

Conducted solar concentrator and receiver optimisation using Monte Carlo Simulation for supercritical CO2 loop in the SERIIUS project.

Monte Carlo Simulations Solar Energy Optimization

Research Assistant

Masdar Institute of Science and Technology

Aug 2013 - Jul 2015 (1 year 11 months)

Performed optical modelling of a 100 kWth beam-down pilot plant for MI-MIT flagship project CSPonD. Used Matlab to create scheme macro in lisp for TracePro, and validated an optical model using heat flux sensor and CCD camera.

Matlab TracePro Monte Carlo Ray Tracing Optical Modeling Image Processing

Research Engineer

Alstom Hydro R&D

Dec 2010 - Aug 2013 (2 years 8 months)

Performed failure analysis of Pelton turbine and turbine components, combining hydraulic and mechanical analysis for Pelton bucket design optimization. Mapped CFD output pressure to FEA model at different time steps and simplified data transfer procedures.

CFDs FEA Pelton Turbine Failure Analysis Mechanical Analysis Telemetry