Admin, Ingestion & Analytics Backend for an Enterprise Chatbot Platform

Overview

The control plane for an enterprise conversational-AI platform: an admin dashboard plus a substantial serverless backend that handles document ingestion, knowledge-base management, access control, and conversation analytics. This is where customer documents become a searchable RAG knowledge base and where teams monitor how their chatbots perform.

The Challenge

Enterprise chatbots are only as good as the documents behind them and the visibility teams have into real usage. The platform needed a single operations surface to upload and ingest source files, manage projects and permissions, and analyze conversation metrics, backed by enterprise-grade secret management and observability.

What We Built

A Vite + React dashboard (package cascade-file-upload) using Mantine and Radix/shadcn UI, Recharts for analytics visualizations, React Router, and reactfire over Firebase. It is backed by a large Firebase Functions backend organized into clear domains: conversation, metrics, file, storage, project, user, acl, rules, queue, and cvs. The ingestion path parses PDFs (pdf-parse), chunks and embeds documents with LangChain (@langchain/openai, @langchain/community) and stores vectors in Pinecone (@langchain/pinecone), with Elasticsearch in the mix for search/indexing. Enterprise integration runs deep: Okta JWT verification, Azure Key Vault for secrets, Azure Monitor (ingestion + query) for telemetry, a Kubernetes client for infra operations, and Google APIs. Tests are written with Vitest and Testing Library.

Technologies & Approach

React/Vite + Mantine for a productive admin UI; Recharts for metrics; Firebase Functions as the serverless backbone; LangChain + Pinecone + Elasticsearch for the RAG ingestion and retrieval layer. The Azure Key Vault / Azure Monitor / Okta footprint reflects the security and compliance expectations of large enterprise customers.

Outcome / Impact

Provided the operational backbone of the platform, turning raw enterprise documents into governed RAG knowledge bases and giving customer teams dashboards to manage projects, permissions, and conversation analytics.

Capabilities Demonstrated

• Document ingestion and chunking/embedding pipelines (PDF to vector store)
• RAG knowledge-base management at enterprise scale
• Conversation analytics and metrics dashboards
• Role-based access control and project management
• Enterprise secret management and observability (Azure Key Vault, Azure Monitor, Okta)