GenAI Backend Engineer at EY (Ernst and Young LLP) (2024-01 – Present)
Validator Productivity Accelerator (Finance - Modelling Risk)
- Built a production-grade multi-agent GenAI platform using LangGraph enabling deterministic orchestration and stateful workflow execution.
- Developed a Gen-AI based internal tool for MRMG Validators to automate labour-intensive processes like Toll Gating, RFR, Findings and Validation Reports. Reduced manual effort of 1–2 weeks per task to 45–50 minutes using advanced techniques including LangGraph orchestration, BM25 retrieval, hybrid context passing and Cassandra Datastax second-level retrieval.
- Engineered a hybrid BM25 + embedding retrieval pipeline with Cassandra Datastax second-level recall improving relevance by 34%.
- Integrated Knowledge Graphs for table relationship modeling, reducing hallucinations and improving factual accuracy.
- Reduced 50% overall processing time by introducing memoization technique (extract context once and consume as and when needed) in the retrieval.
- Built RAG evaluation suite with RAGAS, tracking grounding, precision and hallucination scores effectively reducing token cost by 23%.
GenAI Backend Engineer at EY (Ernst and Young LLP) (2024-01 – Present)
Text-to-SQL RAG pipeline (Finance)
- Developed an end-to-end Text-to-SQL RAG pipeline to query structured data using natural language. Preprocessed JSON, CSV, Excel files into schema format embedded with text-embedding-3-small and stored in DataStax.
- Architected a 3-stage query workflow: Query Rewriting → Schema Linking → SQL Generation using GPT-4o + LangGraph agents.
- Delivered session-based ingestion via S3 and REST APIs deployed through Jenkins CI/CD and Hydra.
- Improved SQL correctness by 27% using schema-aware prompting and fuzzy-schema relevance scoring.
- Integrated Elastic Search and Big Query through GCP (Google Cloud Platform) to execute the SQLs generated.
- Enhanced performance using prompt engineering (CoT, few-shot, schema-aware prompts). Evaluated pipeline effectiveness using RAGAS for factuality and relevance.
GenAI Backend Engineer at EY (Ernst and Young LLP) (2024-01 – Present)
LUMOS - RAG Application (Finance)
- Built a RAG-based financial analysis system that ingests and summarizes quarterly reports from major banks (Barclays, HSBC, Citi) using hybrid retrieval (BM25 + embeddings) and section-aware chunking for accurate grounded summaries.
- Enabled context-aware insights by integrating custom analyst notes, historical data and user-provided context into the retrieval pipeline, improving relevance and reducing hallucinations.
- Implemented multi-hop follow-up QA using LangGraph agents (query rewriting → document routing → grounded generation), delivering interactive, accurate financial insights at scale.
- The solution runs on a scalable GenAI backend using LangGraph for deterministic orchestration, structured chunking tuned for financial PDFs, vector storage in DataStax/Pinecone, and RAG evaluation using RAGAS. Compliance filters (PII detection, hallucination guardrails) ensure enterprise-grade reliability for banking workloads.
Data Analyst at EY (Ernst and Young LLP) (2019-06 – 2023-12)
- RPA: Secured IA projects worth $8 million by pitching automation products to CXOs and Operational Heads and attained 'I Am Exceptional' Award. Awarded 'KUDOS' for streamlining the Reporting process for the client and it became the benchmark for the team of 250+. Spearheaded a team of 5 Configurators and 2 Business Analysts, reducing 70 FTEs using RPA Platforms. Secured $8M+ IA projects by designing automation-led transformation solutions for senior stakeholders. Led a team of 5 developers & 2 BAs, delivering 70+ FTE savings via RPA optimization. Achieved 2500+ hours/year savings through Alteryx-based automation; trained 40+ engineers toward certification.
- Alteryx: Provided an overall 2500+ hours of benefits through Alteryx Automation. Trained 40+ employees on Alteryx and made them Alteryx Certified developers.
