AI DevelopmentRAG / LLMKnowledge ManagementEnterpriseOrganisation Project

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.

“Staff faced challenges navigating complex institutional policies, leading to missed entitlements and suboptimal decisions — all due to fragmented access and time constraints.”

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.

1

Admin uploads policy documentsPDFs stored in MongoDB via GridFS and simultaneously vectorised into Pinecone for semantic search.

2

Smart query classificationEvery user question is first classified by a trained ML model: Is it a known FAQ, a valid policy query, or an irrelevant question?

3

Dual AI response engineFAQ queries return instant predefined answers; non-FAQ valid queries go through LangChain RAG pipeline — retrieving relevant policy chunks and generating accurate LLM responses.

4

Source documents shown as proofUsers see not just the answer but the exact policy document it came from, building trust and transparency.

5

Feedback loop built inUsers submit complaints or suggestions; admins receive notifications, review feedback, and respond — all within the platform.

6

Policy change notificationsWhen admins update or add a document, users are automatically notified, keeping everyone aligned with the latest policies.

Key Features

What's inside

Two-layer AI: FAQ classifier + RAG chatbot
Policy document management (CRUD)
Multi-turn conversation with chat history
Source document shown with every RAG answer
Real-time policy update notifications
Feedback submission & admin response system
Role-based access control (Admin / User)
Document search & filter system

Tech Stack

Built with

FrontendNext.js 14, Tailwind CSS 3.0

BackendNode.js (ES6), Python 3.11

AI / MLLangChain, RAG, Hugging Face (classification)

Vector DBPinecone (semantic search)

DatabaseMongoDB + GridFS (document storage)

InfrastructureHetzner Cloud Server

DesignFigma (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.

The dual-AI approach means the system first classifies intent using a trained Hugging Face model, then routes to either a fast FAQ engine or a full RAG pipeline — giving users instant answers for common questions while handling complex policy queries with full LLM reasoning and document citations.

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.

This project demonstrates our ability to build enterprise AI systems with custom ML pipelines, document intelligence, multi-role platforms, and real-time notification systems — fully deployed on cloud infrastructure.

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