Dec 2022 – Mar 2025 · 2 yrs 3 mos · Chennai, Tamil Nadu, India · On-site

• Management
• Hired and led a team of four machine learning engineers, fostering a collaborative and growth-oriented environment.
• Event Correlation – AIOps Solution for Telecom Fault Management (DISH Telecom, Claro Colombia, Starhub)
• Designed an AIOps solution to handle 1M+ daily telecom fault events, reducing event volume by 60%.
• Applied PrefixSpan sequential pattern mining, discovering 150+ predictive alarm sequences and improving fault prediction accuracy to 75%.
• Built with Python, NetworkX, Rule Mining, Kedro, Docker for scalable deployment.
• Retrieval-Augmented Generation (RAG) – LLM Product (Viavi Internal Product)
• Developed a RAG-based application integrating Langchain with OpenAI GPT models and FAISS vector store for efficient similarity search.
• Enhanced document processing & chunking pipelines using Langchain loaders, splitters, prompts, and chains to enable context-aware Q&A.
• Delivered as a scalable service with Python, FastAPI, PostgreSQL, Docker, Ollama.
• Anomaly Detection for Telecom KPIs
• Built an unsupervised time-series anomaly detection system achieving 85% accuracy on telecom KPIs.
• Combined Isolation Forest with Holt-Winters forecasting to detect anomalies while maintaining interpretability.

Aug 2020 – Dec 2022 · 2 yrs 4 mos · Bengaluru, Karnataka, India · Hybrid

• Ability to Pay (ATP) Model Refresh – Built an ML-based scorecard to assess affordability and credit risk during onboarding. Optimized feature engineering with advanced SQL aggregations, cutting computation time from ~3 days to ~1 hour. Conducted EDA in Python to detect anomalies and improve robustness against NPAs. Achieved AUC of 0.75 (+2% vs. prior) with stronger separation across key risk percentiles. Ensured stability via PSI and CSI monitoring.
• Personal Loan Propensity Model – Predicted likelihood of customers applying for a loan within 7 days post-contact. Engineered features and segmented customers into High/Medium/Low bands, enabling marketing to focus on top propensity cohorts. Improved conversion rates and optimized resource allocation.
• Second Loan & Loan Amount Prediction – Designed models to predict repeat loan uptake, identifying 16.79% within 0–30 days vs. lower later periods. Integrated financial, fraud, and CIBIL data to improve risk separation. Applied ensemble methods and bucket analysis, achieving up to 93% band accuracy in loan amount predictions.
• Income Estimation – Built an ensemble boosted model (XGBoost with custom objective) to estimate income from CIBIL data. Delivered 70%+ bucket accuracy, enhancing credit limit decisioning and risk management.
• Lender Allocation Model – Optimized allocation of applications across lenders, reducing rejection rates by 26% and minimizing settlement time. Used AWS Clarify to ensure explainability, compliance, and transparency.
• Bank Statement Parser (BSP) – Developed a Python & Flask parser with AWS Textract OCR to automate statement extraction. Achieved 89% accuracy in tabular data and 98.6% in transaction categorization. Streamlined data pipelines with Pandas, supporting downstream credit risk models.

May 2019 – Jul 2019 · 2 mos · Pune, Maharashtra, India

• Introduced Guess.js with Next.js for predictive HTTP/2 server push of JavaScript bundles. It uses user navigation patterns to
• determine which pages are most likely to be visited next, thus enabling machine learning-driven user experiences on the Web.

May 2018 – Jul 2018 · 2 mos · New York, United States

• Improve user experience and incorporated the best SEO techniques in existing Shopify website, thereby increasing daily traffic,
• brand positioning and reduction in loading time of the website from 8 seconds to under 3 seconds.