Open to opportunities

Shubham Ashok Gandhi

@shubham_gandhi

Principal Data Scientist delivering production-ready GenAI, Recommender, and ML systems with measurable product impact and low-latency performance.

India

Message

What I'm looking for

I’m looking for Staff, Lead, or Principal roles where I can stay hands-on with GenAI/NLP/ML, make strong product judgments, and deliver cross-functionally—from experimentation to deployment, monitoring, and measurable impact.

I’m a Principal Data Scientist with 10+ years building production ML, GenAI, NLP, search, recommendation, ranking, fraud, and decision systems. I lead applied AI from problem framing and experimentation through deployment, monitoring, and measurable product impact—especially RAG assistants, recommender systems, low-latency inference, conversational NLP, and risk workflows.

At Tiket.com, I designed and delivered an LLM-powered RAG assistant (embeddings, vector retrieval, prompt design) and improved retrieval accuracy by 16–22% with better query handling and synthetic-data fine-tuning. I reworked inference to cut latency from 18 seconds to under 4 seconds and reduced cost by ~97%, and I consolidated NER into a single spaCy pipeline to move from 450ms to under 180ms.

I’ve also driven marketplace growth with sequential recommender systems and marketplace ML (e.g., improving CTR/GBV), simplified candidate/reranker pipelines while boosting CTR/CVR and lowering infrastructure cost, and improved flight-search autocomplete by reducing the mean rank of correct results from 2.1 to 1.13. Earlier work includes fraud and underwriting risk systems at Khatabook and AutoML/ETL with Spark at Razorthink, and I mentor teams while translating experiment results and trade-offs into stakeholder-ready decisions.

Experience

Work history, roles, and key accomplishments

Current

Principal Data Scientist

Current

Tiket.com

Mar 2023 - Present (3 years 3 months)

Led the design and delivery of an LLM-powered RAG assistant, improving retrieval accuracy by 16–22% and reducing generation latency from 18s to under 4s while cutting cost by ~97%. Built and optimized marketplace recommender/ranking and search systems, improving CTR by 8–22%, CVR by 17%, and reducing SQL query latency from 240ms to 55ms.

LLM Applications RAG Systems Embeddings Prompt Design Recommender Systems

Lead Data Scientist

SALT

Sep 2021 - Mar 2023 (1 year 6 months)

Built conversational NLP workflows to convert free-form user text into structured financial categories for expense tracking. Improved personalization by raising spending-category F1 from 63% to 90%+, and fine-tuned DistilBERT/transformer classifiers to raise expense/income/investment classification F1 from ~13–15% to 65%+.

Financial Text Understanding Fine Tuning DistilBERT

Senior Data Scientist

Khatabook

Jun 2020 - Sep 2021 (1 year 3 months)

Developed a fraud detection system using CatBoost, XGBoost, and feed-forward neural networks with Bayesian hyperparameter optimization, reducing fraud rate from 1.3% to 0.08% with periodic retraining via Airflow. Built underwriting decision workflows, including a business-vs-user classifier achieving 92% precision and a pre-approval flow extending credit to 3,500+ customers in a pilot.

Fraud Detection CatBoost XGBoost Neural Networks Bayesian Optimization Feature Engineering

Data Scientist

Razorthink

Jun 2017 - Jun 2020 (3 years)

Led AutoML platform development using Spark for big-data ETL and automated feature engineering, supporting models including XGBoost, LightGBM, and Logistic Regression. Built a life-insurance cross-sell model achieving 74% precision and developed churn prediction for 200M+ users using usage, payment, and demographic data.

AutoML XGBoost LightGBM Logistic Regression predictive modeling

Trainee Decision Scientist

Mu Sigma

Sep 2015 - May 2017 (1 year 8 months)

Used prescribing-behavior data to cluster 265,000 U.S. physicians to support targeted marketing. Built telecom-based credit scoring for micro-credit expansion and developed a document retrieval engine using LDA, k-means, VSM, and TF-IDF for query relevance ranking.

Clustering Topic Modeling (LDA)TF IDF Vector Space Model (VSM)Document Retrieval Ranking Credit Scoring Predictive Analytics