Experience & Timeline

Building production LLM systems — agentic workflows, RAG pipelines, and multi-agent architectures for financial and enterprise domains.

• Built a financial document intelligence system for Master Facility Agreements and Sanction Letters using Docling extraction + semantic chunking (LangChain) to enable long-context retrieval for 100+ page documents.
• Designed a 2×2 VectorDB retrieval architecture (Working Capital vs Long-Term Finance × MFA vs Sanction Letter) to isolate semantic neighborhoods and reduce embedding interference.
• Implemented agentic orchestration using LangGraph to route multi-step workflows (classification → retrieval → synthesis) with deterministic control flow and error handling.
• Integrated Text-to-SQL using Bedrock Agents to support hybrid structured + unstructured reasoning (contract clauses + financial tables).
• Deployed on AWS (Lambda + Bedrock LLMs) with latency, cost, and output formatting optimizations.
• Implemented MCP server tools in Databricks and deployed as an internal app, enabling secure tool integration for LLM-driven workflows.
• Developed an AI-powered accelerator that analyzes product images and descriptions to deliver pricing intelligence, competitor benchmarking, product discovery, go-to-market recommendations, IP risk assessment, and AI-generated marketing content through a multi-agent architecture.

Applied XAI and statistical modeling to supply chain optimization, plus multimodal AI for document processing.

• Designed a safety stock optimization model (Z-score, AMC, lead time, SD), reducing inventory holding costs and improving reorder accuracy.
• Applied Counterfactual Explanations (XAI) to RFQ classifications, identifying actionable parameter shifts that improved deal conversion rates.
• Built a proof-of-concept with Gemini Pro Vision to extract structured data from scanned vendor invoices, streamlining accounts payable workflows.

Delivered analytics and planning solutions within the SAP ecosystem, optimizing enterprise data pipelines and geospatial analytics.

• Delivered analytics and planning solutions within SAP ecosystem, enhancing forecasting, reporting, and decision workflows.
• Reduced processing time from days to minutes by optimizing data pipelines in SAP BPC.
• Integrated spatial + business + financial data across SAP S4HANA, GIS, and Real Estate modules for geospatial analytics.

ML model on 8.9M-device Microsoft Malware dataset with GDPR-compatible explainability for cybersecurity applications.

• Trained ML model on Microsoft Malware dataset (8.9M devices) achieving ~66.7% accuracy and ~66.4% recall, aiding proactive mitigation of data breaches and financial risks.
• Applied EBM, SHAP, LIME for interpretability, producing local + global explanations with uncertainty for regulatory + trust alignment (GDPR-compatible design).
• Engineered features from 4.1GB (83-field) dataset, optimizing memory usage by ~50% via columnar transformations in Pandas/Polars.

Built a multimodal price discovery agent with RAG + tool-calling for 100K+ SKUs.

• Built multimodal price discovery agent using Gemini Pro Vision, achieving ~97% accuracy on internal test set.
• Designed scalable RAG pipeline + VectorDB + tool-calling agent for 100K+ SKUs, reducing manual lookup effort by 50%.
• Implemented fallback logic: queries VectorDB for known items and dynamically falls back to search APIs for unknown products.
• Developed Selenium scrapers for product data acquisition and validation.

• Added usage examples for f_regression() to improve documentation usability — PR #28104
• Added usage examples for silhouette_score() to improve documentation usability — PR #28125