Ashish Kumar Soni

About Ashish Kumar Soni

Data Scientist with 2+ years of experience building production ML systems in banking and fintech. Proven track record developing recommendation engines, fraud detection models, and risk monitoring systems that deliver measurable business impact. Expert in end-to-end ML pipeline development from data ingestion to model deployment, with strong stakeholder management and cross-functional leadership experience. KEY ACHIEVEMENTS:• COMPASS - Customer Recommendation Engine: Built LLM-based system using Edge LLM compute, PySpark, and AWS analyzing 500K+ customer profiles. Achieved 27% lift in cross-sell conversions and reduced mismatched recommendations by 65%• Address Verification API: Engineered ML validation system processing 11M+ addresses with 96.8% accuracy, reducing loan rejections by 67% and saving ₹2.8+ crores quarterly• LoanGuard - Vehicle Loan Risk Model: Developed early warning system for 600+ crore portfolio with 96% F1 score, reducing early-stage delinquency from 7.8% to 3.2% RESEARCH & RECOGNITION:• Distinguished Paper Award at I4AM 2024 Conference, IISc Bangalore for \"A Digital Twin for Extruded Layer Qualification of 3D Printed Concrete\"• 1st Place - Tab Hack 2.0 Hackathon (IITG.ai) among 82 participants• 13th Place - American Express Campus Super Bowl hackathon (99.66 percentile) TECHNICAL EXPERTISE:Programming & ML: Python, SQL, PySpark, XGBoost, scikit-learn, TensorFlow, PyTorch, NLP, LangChain, Mistral-7B, GPT-4 & 5Cloud & Infrastructure: AWS (EMR, Lambda, S3, Athena, EC2), MySQL, PostgreSQL, SAS StudioAnalytics & Visualization: Power BI, Streamlit, Tableau, MLOps & CI/CD pipelinesSpecialized in scalable analytics solutions using cloud infrastructure and advanced statistical modeling. Let\'s connect to explore opportunities in AI and data science!

Experience

1. Manager Analytics

   AU SMALL FINANCE BANK

   Sep 2023 — Present · Jaipur, IN

   Developed and deployed 3 major ML systems in production: customer recommendation engine (COMPASS), address verification API, and vehicle loan risk model (LoanGuard), collectively handling crores in banking operations.• COMPASS - Customer Recommendation Engine: Built LLM-based recommendation system using Edge LLM compute, PySpark, and AWS analyzing customer profiles. Integrated Bureau credit data and behavior analytics via SAS Studio. Achieved XX% lift in cross-sell conversions and reduced mismatched recommendations by XX%. Deployed via Streamlit for 200+ relationship managers, increasing products per customer by XX%.• Address Verification API: Engineered ML validation system processing XXM+ addresses with 9X% accuracy across Home, Personal, and Business Loans. Enhanced precision to 9X% using fuzzy matching (FuzzyWuzzy, Levenshtein) and NLP. Improved collection success rate from XX% to YY%, reducing loan rejections by XX% and saving ₹X+ crores quarterly.• LoanGuard - Vehicle Loan Risk Model: Developed early warning system using XGBoost, PySpark, and AWS EMR/Athena for crore portfolio. Achieved 96% F1 score identifying defaulters before delinquency. Tracks XX+ indicators across loans, captured XX% of likely defaults pre-delinquency. Reduced early-stage delinquency over 8 months.• Led stakeholder presentations to senior leadership, conducted technical training for analytics team, and established ML model monitoring frameworks using Power BI dashboards for real-time portfolio health tracking.

Education

• Indian Institute of Technology, Guwahati

  Master of Technology - MTech, Robotics And Artificial Intelligence

  2021 — 2023

• M.B.M. Engineering College, Jodhpur

  Bachelor of Engineering - BE