Data Scientist - HCL TECH Pvt Ltd - Hyderabad
(2024-08)
Architected an Agentic AI Legal Assistant using Multi-Agent RAG architecture to automate legal research, statute interpretation, case summarization, compliance validation, and evidence retrieval workflows across 358 IPC sections.
- Built scalable legal AI pipelines using GPT-4, LangChain, and Transformer-based NLP models for contextual legal Q&A, grounded response generation, and intelligent document analysis.
- Coordinated specialized AI agents for legal research, summarization, and evidence retrieval to support multi-step legal reasoning and workflow automation.
- Generated semantic embeddings using OpenAI text-embedding-3-large and implemented Qdrant Vector Database with hybrid semantic retrieval and cosine similarity search for high-accuracy contextual retrieval.
- Optimized retrieval quality using RecursiveCharacterTextSplitter, metadata filtering, Multi-Query Retrieval, and BERT-based semantic re-ranking, improving contextual grounding and retrieval relevance.
- Evaluated RAG pipeline performance using RAGAS metrics, achieving 0.87 Faithfulness, 0.89 Answer Relevance, 0.88 Groundedness, and 0.81 MRR while reducing hallucinated responses through retrieval and prompt optimization.
- Applied Pydantic schema validation and structured response enforcement, reducing formatting inconsistencies by 25% and improving reliability of AI-generated legal outputs.
- Reduced manual legal research and document review effort by 40–50%, accelerating contract analysis and case review workflows through AI-driven retrieval and summarization automation.
Jr. Data Scientist - Mphasis - Hyderabad
(2023-03 - 2024-07)
Delivered an AI-powered Banking Chatbot using GPT-3/GPT-4 to streamline customer support, transaction assistance, account services, and personalized banking interactions.
- Enabled accurate query understanding through BERT-based intent classification and entity recognition pipelines, improving extraction accuracy by 18% for complex banking requests.
- Facilitated context-aware and multi-turn conversations using retrieval-augmented response generation and reinforcement learning-based dialogue management, improving recommendation relevance by 22%.
- Secured real-time banking workflows through JWT-based authentication, role-based access control, and REST API integrations while maintaining 99.5% service uptime.
- Optimized conversational routing, response handling, and prompt workflows to support high concurrent traffic with low-latency interactions; achieved 92–95% intent accuracy and 89–92% entity recognition F1-score.
- Improved operational efficiency by automating 60%+ of routine customer queries, reducing response time from minutes to seconds, and sustaining 93.7% customer satisfaction post-deployment.
