Anil Jhanwar
Agentic AI & Data Science Leader | LLM · RAG · Multi-Agent Systems | Credit Risk · Fraud Detection · Collections AI | BFSI · Fintech | $10M+ Impact | IIT Roorkee · CMU
Gurugram
9+ years of work experience
Jul 20233 yrs 2 mosPresent
Gurugram, Haryana, India
Head of Data Science | Agentic AI · Credit Risk Modeling · Collections AI · LLM Voice Agent | IndusInd Bank
AI Voice Agent for Loan Recovery | Agentic AI · LLM · STT/TTS · Multi-Agent | $4M Impact
• Led end-to-end design and production deployment of an AI-powered voice agent for collections using LLM-driven dialogue management, speech-to-text (STT) and text-to-speech (TTS), enabling automated, human-like borrower interactions across delinquent segments.
• Integrated ML-based risk segmentation with adaptive conversational strategies. Improved early-stage recovery rates by 8% via higher promise-to-pay (PTP) conversion — $2M impact in first quarter. Scaled to 3 languages (EN/ID/TH), 62x volume increase, 97% cost reduction ($4.50 → $0.12/call).
Credit Card Application Scoring | Credit Risk Modeling · Credit Scorecard · XGBoost · WOE/IV | ₹800M Impact
• Owned end-to-end credit scorecard development using XGBoost, logistic regression, WOE/IV feature engineering, and AutoML. Bank achieved 4% higher customer acquisition from revamped application scoring while maintaining identical risk profile. ₹800M uplift.
• Delivered Model Risk Management (MRM) documentation, bias audit, and governance framework — reduced MRM validation cycle by 50%.
Collections AI | Default Prediction · Behavioral Scoring · TransUnion · ₹500M Impact
• Built credit card default prediction and early settlement model in collaboration with TransUnion using behavioral scoring, feature engineering, and threshold optimization. Added ₹500M in collections, reduced NPAs.
Banking Data Integration | Portfolio Analytics | Cost Efficiency ₹100M
• Identified systemic gap from isolated acquisition and collections data silos. Built real-time reverse model feedback integration — improved collection efficiency and reduced NPA losses by ₹100M.
Jul 20179 yrs 2 mosPresent
My AI & Data Science Project portfolio: https://anil10iitr.github.io/
Please visit respective GitHub Links to view the actual code and analysis.
(i) Predictive Model for Seizure Detection in Epileptic Patients| PCA | XGBOOST | Randomized Grid Search
• Developed a very effective XG Boost model on the high dimensional brain activity record (EEG) data for epileptic patients. Devised a robust approach to handle the computational challenges because of dimensionality of data. Performed PCA followed by several baseline model to spearhead the right approach.
• Eventually employed Randomized Grid Search to get optimal hyperparameters which delivered an accuracy of 97% which was critically higher than baseline model. Further, the Random Grid search cut down the computational time by 50%.
(ii) Future Sales Prediction of a Grocery Store | XGBoost | ANN | Ensemble Modeling
• Built an ensemble of XG Boost and ANN model to get the optimal performance score for predicting the sales volume. Deftly handled extremely large volume of data (1 billion rows) by building the base model on random sample of a smaller fraction and then built the final model with optimal hyperparameters on whole dataset.
• This practical approach to the huge dataset helped score in top 10% on leaderboard with a short span of 2 weeks using the basic computing resources. Nonetheless, it helped in avoiding overfitting of the model significantly.
Oct 20212 yrs 1 moOct 2023
Pittsburgh, Pennsylvania, United States
Data Science Lead | AI CoE Builder | LangChain · RAG · NLP · Production ML | Fortive Corporation
RAG-Powered Internal Knowledge Assistant | LLM · FAISS · NLP · 5,000+ Users
• Built a Retrieval-Augmented Generation (RAG) pipeline over Fortive’s engineering and product wiki corpus using LangChain, FAISS vector store, and OpenAI embeddings. Benchmarked on 30 domain-specific queries: base GPT-4o achieved 58% accuracy vs 85% for the RAG-enabled system — a 45% relative improvement on internal terminology where base LLMs hallucinate.
• Adopted by 5,000+ employees across the organisation. Reduced new-hire onboarding research time from days to minutes. Reusable architecture across policy Q&A, technical documentation, and compliance workflows.
Customer Churn Prediction | AutoML | Production ML | Dataiku | $1M Savings
• End-to-end predictive model development and production deployment for customer churn prediction using AutoML on Dataiku platform. Enabled retention team to achieve $1M in additional savings and 30% YoY reduction in customer churn.
• Built fully automated MLOps pipeline with zero human intervention required for monthly scoring cycles.
Feb 20183 yrs 9 mosOct 2021
Greater Pittsburgh Region
Data Science Project Manager | Fraud Detection · Feature Engineering · NLP · Credit Analytics | EXL
Fraud Detection Model | Credit Card Portfolio · Recall Optimization · $3M Savings | Top-10 US Bank
• Built a fraud detection model for a leading US bank credit card portfolio. Prioritised recall over precision — minimised false negatives, reducing false negative rates by 5% and translating into significantly lower fraud losses (~$3M saved annually).
• Led data integration from multiple unconnected source systems, enabling richer feature engineering and stronger fraud signal extraction. Maximised prediction performance across card-not-present and card-present fraud segments.
Customer Analytics | Churn Prediction · AutoML · Dataiku DSS | $1M Savings
• Delivered end-to-end churn prediction and prevention model for a major financial services client using AutoML on Dataiku DSS with advanced feature engineering. Retention team achieved $1M additional savings and 30% YoY churn reduction.
• Built fully automated production ML pipeline with zero human intervention for monthly scoring cycles.
Jul 20177 mosJan 2018
Greater Philadelphia Area
Client: City of Philadelphia
• Built a predictive model enabling the city government to take timely decisions for safety of vulnerable children. The Model predicts the level of risk for children, with 15% better accuracy than human judgement, and substantially reduces the child fatalities.
• Proposed strategic saving plan of $ 8 million by developing supply-chain optimization model for vehicle replacement. Used Excel based solver platform to build optimization model by analyzing large sample of vehicle expense data.
Anil’s AI Stack
Education
Master's degree, Data Analytics,Management & Public Policy
Carnegie Mellon University
20152 yrs2017
Bachelor of Technology - BTech
Indian Institute of Technology, Roorkee
20064 yrs2010
Join Anil and 10k+ other professionals on
Tal is the app where bosses meet engineers & builders directly, without middlemen
Download Tal Boss