SHIKHARPANDEY

Designed and developed a Retrieval-Augmented Generation (RAG) chatbot over an enterprise healthcare knowledge base, using LangChain agent-based orchestration on FastAPI to route queries across multiple data sources. Built a MongoDB vector store with embedding & retrieval pipelines and reranker models (Voyage, Cohere) to boost answer relevance; added token-streaming responses with source citations. Implemented secure document ingestion pipelines (PDF / DOCX / PPTX / XLSX) with metadata extraction, indexing and Azure Blob Storage, plus role-based access filters and chat audit analytics. Built the responsive React.js chatbot UI for seamless, context-aware user interactions. Built a NL-to-query engine (FastAPI + LangGraph + LangChain) converting plain English into MongoDB aggregations, SQL Server T-SQL, and Databricks SQL, powered by Azure OpenAI GPT-4o. Engineered three distinct agent architectures: a ReAct SQL agent (SQLDatabaseToolkit), a two stage LangGraph MQL pipeline (rewrite generate), and a multi-node LangGraph charting agent auto-selecting among 19 Plotly chart tools. Integrated Databricks Genie (OAuth) and hardened LLM query execution with sandboxed evaluation, operator allow-listing and a full query audit trail (latency, row counts, feedback). Built the React + TypeScript + PrimeReact BI UI secured with Azure AD (MSAL); shipped via Docker. Built a production Teams bot (Python Bot Framework + async aiohttp) surfacing the NL2Query/RAG backend with Adaptive Cards (native tables, base64 chart images), feedback loops, and Azure AD identity; deployed to Azure App Service. Added categorized follow-up query suggestions as Teams SuggestedActions and refactored the bot to drop a legacy SQL dependency, cutting startup latency. Delivered a multi-tenant healthcare training platform: NestJS 11 + MongoDB + Redis (BullMQ) backend and React 19 + Vite + MUI + Redux Toolkit frontend. Implemented fine-grained RBAC across 6 roles with market-scoped data access (CLS + Mongoose scope plugin preventing object-level auth bypass), Azure AD auth, delegation grants, digital sign offs, and PDF compliance certificates. Built async pipelines (BullMQ email / notification / auto-assignment queues), scheduled at-risk scans, and Playwright E2E tests; CI/CD to Azure Container Apps. Built sales enablement AML Sanction-Compliance Agent autonomous OFAC/UN/EU pipeline using DeepAgents + LangGraph + LangChain v1; a planner orchestrating 6 specialized sub-agents (ingest parse diff rule HITL finalize), with human-in-the-loop approval via LangGraph interrupt(), a MongoDB $vectorSearch RAG chatbot, audit logging, and LangSmith tracing. Built NL-to-Chart Analytics over MCP FastAPI + LangGraph pipeline turning sales questions into MongoDB aggregations and AntV PNG charts through Model Context Protocol servers; authored a custom async JSON-RPC / SSE MCP client and a query-safety validation layer (operator whitelist, stage/row limits).

Integrated the OpenAI API + RAG into a multi-tenant SaaS for intelligent retrieval and AI-powered responses. Built secure OAuth2 auth with Django REST Framework (GitHub & Google logins) plus a granular RBAC model; designed multi-tenant support with PostgreSQL read-replica optimization and a dedicated authentication microservice. Ran background jobs (bulk updates, notifications) asynchronously with Redis Queue (RQ) & Celery; shipped collaborative object-sharing with Django & Next.js.

Bank of America (via Cisco): Processed large payment-transaction datasets with Python & Pandas for real-time dashboards; optimized Splunk SPL queries to improve indexing speed; built React monitoring dashboards. Airbus (Skywise / Palantir): Wrote complex SQL + Pandas pipelines to clean and join data for maintenance inspection dashboards; automated web-dashboard testing with Selenium.