Desai

Software Engineer with 1+ year of experience building and deploying production-grade AI systems in machine learning and backend engineering. Hands-on experience developing NLP pipelines, LLM-powered applications, and RAG architectures using Python, FastAPI, and Docker — reducing manual review effort by 70% across 10,000+ documents. Proficient in designing scalable REST APIs, multi-agent systems with LangGraph and LangChain, and large-scale data pipelines with Apache Spark.

Experienced in A/B testing, model evaluation, and LLM fine-tuning (LoRA/PEFT), with deployments on Azure and AWS.

Associate Data Scientist at Straive (2024-09 – 2025-09)

Built and deployed production-grade AI systems for document intelligence and legal domain applications

• 80% reduction in manual review effort across 10,000+ editorial documents — achieved by deploying end-to-end LLM pipelines with PyMuPDF OCR ingestion, section-aware chunking, BM25 + dense vector hybrid retrieval, and cross-encoder reranking integrated via FastAPI microservices
• Eliminated hallucinations in legal domain Q&A by applying chain-of-thought (CoT) prompting, few-shot examples with domain-specific legal terminology, structured JSON output generation, and grounding layers that cited retrieved source passages before generating answers
• Designed production-grade guardrails and validation layers for AI-generated outputs — including input sanitization, confidence thresholding, output schema enforcement, and fallback routing — reducing compliance risk across automated document review workflows
• Enabled dynamic query routing by integrating OpenAI function calling into FastAPI microservices, allowing structured user queries to be dispatched to the correct tool, retrieval pipeline, or LLM execution path without manual mode selection
• Built and deployed containerized ML microservices using Docker and FastAPI, enabling modular ingestion, retrieval, and generation workflows with seamless integration into existing document management systems — ensuring reproducible, production-ready deployments
• Established model evaluation and QA frameworks tracking precision, relevance, and latency metrics across AI pipelines — applying offline evaluation strategies and prompt engineering to continuously improve LLM accuracy for domain-specific terminology
• Engineered scalable batch data pipelines processing structured and unstructured data at scale — designing ETL workflows for document ingestion, transformation, and feature extraction to support downstream ML model training and inference
• Collaborated cross-functionally with product managers, backend engineers, and domain experts to translate complex business requirements into well-defined ML solutions with clear success metrics and iterative delivery under Agile/Scrum workflows
• Implemented experiment tracking and model versioning practices across development cycles — maintaining reproducibility of ML experiments and enabling systematic comparison of retrieval strategies, chunking configurations, and prompt templates

Under Graduate in Computer Science and Engineering – Amrita School of Engineering (2020 – 2024)