Aakash Agarwal

About Aakash Agarwal

Experienced Data Scientist with strong math background and a demonstrated history of working in the Fintech, consumer electronics and Telecom industry. Skilled in Deep Learning, Pytorch, CV, NLP, Python, Keras, TF and Data Structures. Strong engineering professional with a B.Tech from

Experience

1. Senior Director Data Science

   Paytm

   Feb 2024 — Present · Gurugram, IN

   Leading the development of next-generation Voice AI for Paytm Enterprise Voicebots, following a phased architecture roadmap from Speech-to-Reason (S2R) toward fully end-to-end Speech-to-Speech (S2S) systems- Currently building a Speech-to-Reason model inspired by Ultravox, using a Whisper encoder, projection layer, and an enterprise-fine-tuned LLM to reason directly over speech embeddings and generate responses; this serves as a stepping stone toward future direct S2S model deployment- Led the research and development of a Merchant Foundation Model trained on billions of Paytm transactions, generating rich merchant vector embeddings that serve as reusable representations across multiple business use cases- Leveraging Foundation Model embeddings, built and productionized downstream models for Merchant Churn, Merchant Risk, and Merchant Lending (credit underwriting & recovery), achieving 10–15% improvement over existing production models across key business metrics and reducing end-to-end development time to under 2 weeks per use case- Developed Merchant risk model to flag risky merchants and decrease the workload of ops team by 50% and additional improved the risky merchant capture rate by 20%- Developed Multi-modal Vision based models which takes merchant shop photos as input and predictwhether the shop is movable, merchant category code and additional parameters. It helped the ops team to fasten onboarding by 12% while reducing cost by 17% in the first month itself.

Education

• St Paul School

  Physics, Chemistry and Maths

  1999 — 2009

• International Institute of Information Technology Hyderabad (IIITH)

  B.Tech, Electronics and Communication

  2009 — 2013

Skills

• Natural Language Processing
• Machine Learning
• Cadence Virtuoso
• Deep Learning
• Vlsi
• Recommender Systems
• C
• Field-Programmable Gate Arrays (Fpga)
• Algorithms
• Fpga
• Verilog
• Cadence
• Vhdl
• Html
• Machine
• Tanner Eda
• Very-Large-Scale Integration (Vlsi)
• Signal Processing
• Tensorflow
• Testing
• Java
• Matlab
• Nltk
• Modelsim
• Data Structures
• Keras
• C++
• Nlp
• Altera Quartus
• Microcontrollers
• Python