Experience & CV
Experience
My work across applied AI, machine learning systems, and research.
Building, evaluating, and deploying AI.
Amazon
Applied Scientist
- Built grounding guardrails for a production multi-agent root-cause analysis (RCA) system on AWS Bedrock, enforcing 70+ validation checks and reducing false-positive escalations by 72%
- Designed a reference-based evaluation framework using 3 LLM judges to benchmark agent diagnoses against 6K+ expert-resolved tickets, improving diagnosis accuracy by 80%
- Developed an evaluation and observability layer across 8+ services for Amazon Robotics’ RFID dangerous-goods detection system, reducing debugging time by 40% and supporting $70M+ in annual savings
Walmart Global Tech
Data Scientist III, Applied AI
- Developed an agentic RAG assistant enabling merchandising teams to query retail data in natural language, validating relevance and groundedness to reduce manual SQL work by 40%
- Fine-tuned and benchmarked LLMs across prompting and retrieval configurations, improving answer quality by 18% over baseline and reducing response latency by 2 seconds across 5 use cases
- Engineered Spark and BigQuery pipelines processing 4+ TB/day of training and evaluation data, improving downstream model accuracy by 30% and serving predictions to 200+ analysts via FastAPI
ZeoAuto (Y Combinator)
Software Engineer (Backend & ML Systems)
- Deployed serverless Python services on AWS Lambda for real-time ML routing inference, handling 15K+ daily requests with 99.9% uptime and production monitoring
- Integrated ML routing models into the Zeo Fleet platform, reducing delivery times by 20%, and optimized FastAPI and Docker inference APIs to cut end-to-end latency by 30%
AI Institute, University of South Carolina (AIISC)
AI Researcher
- Developed multimodal data pipelines for FACTIFY 3M, a large-scale fact-verification benchmark, combining 1M+ claims from 8 sources with 3.5M+ LLM-generated paraphrases and 3.3M+ synthetic images
- Evaluated text-only and multimodal models on 135K adversarial stories, demonstrating 2.4× higher F1 scores for multimodal approaches, contributing to an EMNLP 2023 Main Conference Publication
Indian Institute of Technology Delhi
ML Researcher
- Trained and tuned 5 ML models, including gradient boosting, random forest and KNN, with cross-validation and grid search to predict the remaining useful life of NASA turbofan jet engines with 83% accuracy
- Reduced the feature set from 27 to 10 sensor signals using SHAP and partial dependence analysis to identify failure drivers, resulting in a Springer International Conference Publication at IISc Bangalore
Academic foundations.
The University of Texas at Dallas
M.S. Computer Science (Applied Machine Learning)
Richardson, TX
Birla Institute of Technology and Science (BITS), Pilani
Dual degree: M.Sc. + B.E.
Pilani, India
Research contributions.
FACTIFY3M: A Benchmark for Multimodal Fact Verification with Explainability through 5W Question-Answering
EMNLP · Main Conference
Advancing Predictive Maintenance: A Data-Driven Approach for Accurate Equipment Failure Prediction
Industry 4.0 and Advanced Manufacturing
Drug Discovery Using Artificial Intelligence
International Journal of All Research Education & Scientific Methods
Heart Disease Prediction by Machine Learning Over Big Data from Health Care Communities
International Journal of All Research Education & Scientific Methods
Mathematical Modeling of Simultaneous Nitrification and Denitrification in Bioreactor Systems
National Conference on Sustainable Technologies (STCBS)
Technical toolkit.
Languages
Python, Java, TypeScript, C/C++, SQL, Bash
AI / ML
Multi-Agent Systems, Agent Training & Evaluation, LLM-as-Judge, Fine-tuning, RAG, RL
Frameworks
PyTorch, Hugging Face, Scikit-learn, LangChain, Strands Agents, FastAPI
Systems
Distributed Training (DDP), REST APIs, Spark, BigQuery, Docker, CI/CD
Cloud & infrastructure
AWS (Bedrock, AgentCore, SageMaker, Lambda, S3, IAM, VPC), Kubernetes, Terraform