A colorful, standalone AI-102 portfolio project for document ingestion, retrieval, grounded Q&A, and Azure-ready upgrade paths.

Overview

AI-102 Knowledge Assistant is a standalone portfolio project that demonstrates the delivery pattern behind a modern Azure AI knowledge app:

• ingest support documents and normalize them into chunks
• search indexed content with grounded retrieval
• answer user questions with citations
• apply safety filtering and light redaction
• switch between local and azure runtimes without changing the UI

The sample scenario uses Nimbus Home, a fictional smart appliance brand whose support teams need fast answers about setup, error codes, warranty rules, and escalation criteria.

Why It Stands Out

• Strong AI-102 story: ingestion, retrieval, prompt grounding, and guardrails in one project
• Real demo flow: local MVP works without paid cloud dependencies
• Cloud upgrade path: Azure AI Search, Azure OpenAI, and Azure AI Document Intelligence
• Recruiter-friendly: clean UI, runnable app, tests, docs, and architecture notes

App Preview

Feature Snapshot

Area	Included
Local web app	http.server UI and API
Local retrieval	Markdown ingestion, chunking, JSON index
Grounded answers	Deterministic local answer composer
Guardrails	Prompt blocking and contact-data redaction
Azure runtime	Optional search + generation with Azure services
Document import	Optional Document Intelligence pipeline for PDF and image files
CI	GitHub Actions unit-test workflow

Runtime Modes

Mode	What It Uses	Best For
local	Python standard library only	Fast demos, interviews, offline MVP
azure	Azure AI Search, Azure OpenAI, optional Document Intelligence	Real cloud showcase and AI-102 alignment

Architecture Map

Local Component	Azure AI Equivalent
Markdown ingestion and chunking	Azure AI Document Intelligence extraction pipeline
Local JSON index	Azure AI Search index
Deterministic answer composer	Azure OpenAI grounded generation
Prompt safety rules	Azure AI Content Safety or app-side guardrails
Local web app	App Service, Container Apps, or Functions hosted API

More detail: Architecture Notes

Quick Start

Start the local app:

cd path/to/ai102-knowledge-assistant
python3 -m app.server

Open:

http://127.0.0.1:8080

Rebuild the local index:

python3 scripts/rebuild_index.py

Run tests:

python3 -m unittest discover -s tests

Azure Mode

Install optional cloud dependencies:

python3 -m pip install -r requirements-azure.txt

Create a local .env file from .env.example, then set:

• KNOWLEDGE_ASSISTANT_MODE=azure
• AZURE_OPENAI_ENDPOINT
• AZURE_OPENAI_API_KEY
• AZURE_OPENAI_CHAT_DEPLOYMENT
• AZURE_SEARCH_ENDPOINT
• AZURE_SEARCH_API_KEY
• AZURE_SEARCH_INDEX_NAME

Optional for raw file import:

• AZURE_DOCUMENT_INTELLIGENCE_ENDPOINT
• AZURE_DOCUMENT_INTELLIGENCE_API_KEY
• AZURE_DOCUMENT_INTELLIGENCE_MODEL

Import PDF or image files:

python3 scripts/import_with_document_intelligence.py --input-dir data/raw_documents

Sync the local chunk set into Azure AI Search:

python3 scripts/sync_to_azure_search.py --rebuild

Full setup guide: docs/azure-setup.md

Refresh Preview Artifact

To regenerate the GitHub preview screenshot from the live local app:

python3 -m app.server
npm install
npx playwright install chromium
npm run capture-preview

Demo Questions

• How do I pair the oven with Wi-Fi?
• What should I do when the BrewMaster Mini shows E12?
• How long is the warranty period?
• When do I escalate a premium support case?
• Show me the support passwords.

The last question intentionally demonstrates the safety policy.

Project Structure

• app/server.py
• app/runtime.py
• app/knowledge_base.py
• app/azure_integration.py
• scripts/rebuild_index.py
• scripts/sync_to_azure_search.py
• scripts/import_with_document_intelligence.py
• docs/architecture.md
• docs/azure-setup.md
• docs/demo-script.md

Interview Talking Points

• How retrieval grounding reduces hallucinations
• Why chunking strategy matters for support and policy documents
• How the same app can support both a lightweight MVP and a cloud-backed architecture
• Where Content Safety, telemetry, authentication, and feedback loops would fit in production

License

Released under the MIT License.