Bengaluru, Karnataka, India
I am Mani Ratnam. Recently I was a Sr Analytics Scientist at a fintech building ML models, ETL pipleines, running A/B tests. I hold a bachelors degree in Petroleum engineering from IIT Dhanbad. Ever since I discovered a knack for data, I've been learning SOTA techniques, finding creative ways to solve problems and push beyond whats possible. Experienced in Traditional ML as well as new age LLMs and Agentic AI. I am currently trying to solve for the optimal buying strategy in e-commerce B2C as a side hustle that can revolutionize shopping online thus I call it project 'Nobius'(pronounced 'No BS').
Sr. Analytics Scientist | Risk & Data Science Overall generated Intelligence to Collections/Operations, Product and Marketing using ML and guided those teams in taking the right path in their ops. Led analytics and ML-driven collections strategy for a $50M lending portfolio, focusing on repayment uplift, delinquency reduction, and cost efficiency of small ticket instant loans for the underserved indian B2C credit segment. Built and deployed pre-due and post-due XGBoost models to predict on-time and 30+ DPD outcomes using bureau data and behavioral features, with evaluation via AUC, KS, and risk-based binning for production rollout. Operationalized risk models across risk-segmented communications and tailored communications based on channel engagement metrics, AI voice agents, and incentive programs, delivering +6% uplift in on-time payments and +4.5% improvement in 30+ DPD recovery rates, while reducing monthly marketing costs by 11% through rolling cost optimization. Drove experimentation and A/B testing in Underwriting, resulting in +3% improvement in on-time payments. Built fraud and concentration risk detection frameworks using employer and pin code profiling, and built an NLP frame work using Regex to help stardardize names across Indian government IDs like PAN/ Aadhaar /Bank Acc name. POCed a GenAI-powered voice collections agent (STT–LLM–TTS) with Indic language support, integrating RAG and VAD for scalable, automated calling.
Built Delinquency models for various countries, used tree based and boosting based algorithms to predict defaulters which helped optimize collection teams' efforts and increased collections value by 30% - Created similarity checking frameworks on conversational data for KPIs for the retail clients. Used BERT based pretrained models to generate embeddings later changed the tech stack to GUSE Embeddings and FAISS. Implemented teh retireval part of RAG from scratch even before GenAI was a buzz word - Rewritten custom deep learning architectures of axquired CXM startup in Pytorch after deciphering the code base. - Taken care of the NLP pre-deployment pipelines for the insurance giant AFLAC, using Apache airflow, created airflow pipelines; fixed them when they failed.
- Data strategist to create new products in the lines of research and advisory. - Built unsupervised sentiment analysis, Topic Modeling/clustering frameworks for social media data(Twitter, fb, forums, blogs, articles ) - Helped the researchers by automating text cleaning and dataset structuring using python. - Conducted secondary research on markets like AI, 5G, observability, MlOps etc. - Prototyped a python frame work integrated with UiPath for image analytics. The framework scrapes the images across web and performs OCR (Optical Character Recognition) using Google API then transforms the image into text and creates a dataset that will be further used to perform various data science techniques relevant to the case/problem.
-Created crawlers to gather data using scrapy across the web. -Cleaned and organized the data from unstructured data sources.