Aman Agarwal

About Aman Agarwal

Experienced Senior Data Scientist with 8 years in the IT industry, specializing in machine learning, deep learning, NLP and Generative AI. Proven track record of leading data-driven projects to achieve significant business outcomes, collaborating effectively with international teams, and deploying production-ready models. I strongly believe that with these new advances in technology like Deep learning and computation power that we have, we can bring changes to a lot of things in our daily life which makes our life easier.

Experience

1. Lead Data Scientist

   Publicis Sapient

   Jul 2024 — Present · IN

   Project: Enterprise NL2SQL Platform•Architected a production-grade Multi-Agent LLM-based NL2SQL system, reducing query turnaround from minutes to 10 seconds and improving execution accuracy by 35%.•Designed hybrid Schema + Value Retrieval using Elasticsearch, improving SQL grounding and reducing regeneration cycles by 25%.•Built real-time streaming inference using Async Python, FastAPI, reducing latency by 50% under concurrent enterprise workloads.•Developed scalable microservices deployed on Kubernetes (multi-pod architecture) with checkpointing, retries, circuit breakers, achieving 99.9% uptime.•Optimized LLM cost and cloud infrastructure using token caching, object storage strategies, and autoscaling, reducing inference spend by 40%.•Implemented enterprise-grade security, governance, and access controls for multi-tenant deployments.•Integrated solution into RAG pipelines, vector databases, and metadata-driven workflows.•Collaborated with global product, governance, and engineering teams to deliver compliant enterprise AI systems.Project: LLM-Based Enterprise APIs (Attribute & Summary Extraction)•Designed and deployed production-grade FastAPI for structured attribute extraction and document summarization (extractive & abstractive) using Large Language Models (LLMs).•Built a flexible schema-driven validation framework supporting multiple data types, configurable business rules, and structured JSON outputs for downstream automation.•Implemented dynamic prompt orchestration, contextual grounding, and output validation to ensure high accuracy and enterprise reliability.•Developed section-wise summarization pipelines with strict schema enforcement for large PDF and text document processing.•Integrated async processing, logging, monitoring, and robust error-handling mechanisms to support scalable, multi-environment deployments.•Optimized token usage and inference costs, improving efficiency for high-volume enterprise document workflows.

Education

• Dr. A.P.J. Abdul Kalam Technical University

  Bachelor of Technology (BTech), Electrical, Electronic and Communications Engineering Technology/Technician

  2011 — 2015

• Analytixlabs

  Data science using SAS and R

  2017

• Indian Institute of Technology, Delhi

  Executive Program, Data Science & Machine Learning