DocuMind — AI-Powered Policy & Knowledge Assistant
An intelligent internal assistant built for a large organisation, enabling staff to instantly query complex policy documents, compliance rules, and institutional knowledge — powered by RAG and LLM technology.
6
Core modules built
3-tier
System architecture
RAG
AI engine powering responses
2-layer
AI query classification
The Problem
Hundreds of policies, zero easy access
Large organisations often have hundreds of internal policies — scholarship rules, compliance guidelines, leave procedures, grant regulations — scattered across documents nobody reads. Staff waste hours searching for answers, often miss entitlements, and overwhelm support offices with repetitive questions.
What We Built
Full-stack AI knowledge platform with dual-AI engine
A full-stack AI assistant platform where admin teams upload and manage policy documents, and staff interact with a smart chatbot that answers policy questions instantly — with source documents shown as evidence. Two-layer AI architecture: fast FAQ matching for common questions, and full RAG + LLM generation for complex queries.
Admin uploads policy documents — PDFs stored in MongoDB via GridFS and simultaneously vectorised into Pinecone for semantic search.
Smart query classification — Every user question is first classified by a trained ML model: Is it a known FAQ, a valid policy query, or an irrelevant question?
Dual AI response engine — FAQ queries return instant predefined answers; non-FAQ valid queries go through LangChain RAG pipeline — retrieving relevant policy chunks and generating accurate LLM responses.
Source documents shown as proof — Users see not just the answer but the exact policy document it came from, building trust and transparency.
Feedback loop built in — Users submit complaints or suggestions; admins receive notifications, review feedback, and respond — all within the platform.
Policy change notifications — When admins update or add a document, users are automatically notified, keeping everyone aligned with the latest policies.
Key Features
What's inside
Tech Stack
Built with
Frontend — Next.js 14, Tailwind CSS 3.0
Backend — Node.js (ES6), Python 3.11
AI / ML — LangChain, RAG, Hugging Face (classification)
Vector DB — Pinecone (semantic search)
Database — MongoDB + GridFS (document storage)
Infrastructure — Hetzner Cloud Server
Design — Figma (UI prototyping)
What Made This Complex
Not just a chatbot
Most chatbots just do one thing — answer from a knowledge base. DocuMind required a significantly more sophisticated architecture.
This combination of ML classification + LangChain RAG + MongoDB GridFS + Pinecone vector search — all on a production cloud server — represents a fully enterprise-grade AI knowledge system, not a demo or prototype.
The Outcome
Hours of searching replaced by instant answers
DocuMind replaced hours of manual policy searching with instant, accurate, source-backed AI answers. The organisation's support team saw a significant reduction in repetitive policy queries. Staff could now self-serve answers to complex compliance and procedural questions — with confidence, because the system always showed the source document behind every response.
Ready to build?
Want something like this?
Tell us about your project. We'll come back with a custom scope and proposal — no pressure.
Book a Free Discovery Call