InventoryIQ

A Microsoft Foundry agent for IT asset & datacenter reasoning, grounded in Foundry IQ over a live Azure AI Search knowledge source.

Built for the Microsoft Agents League Hackathon — Enterprise Agents track, June 4–14 2026.

What it does

Ask your IT estate anything in natural language and get a grounded, cited answer:

• "Which servers in Building B / Floor 2 raised critical alerts in the last 24 hours?"
• "Show laptops assigned to the Finance department that haven't checked in for 7 days."
• "Which assets on the DC-East blueprint are in the cold-aisle zone?"
• "Which Windows assets are running unsupported OS versions, grouped by location?"
• "Draft a procurement ticket for replacement drives based on this quarter's failure pattern."

Every answer is grounded in Microsoft Foundry IQ over an Azure AI Search knowledge source mirrored from the InventoryMapper .NET 10 asset platform, with citations back to the source row.

Architecture

 ┌─────────────────────┐    EF Core / SQL Server
 │  InventoryMapper    │──────────────┐
 │  (.NET 10 API)      │              │
 └─────────────────────┘              ▼
                              ┌───────────────────┐
                              │  grounding/       │  ingestion script
                              │  scripts/         │  (Python)
                              └─────────┬─────────┘
                                        ▼
   ┌──────────────────────────────────────────────────────┐
   │  Azure AI Search index  inv-index                    │
   │   685 docs: asset + alert                            │
   │   1536-dim vector embeddings (text-embedding-3-small)│
   │   semantic ranking enabled                           │
   └─────────────────────────┬────────────────────────────┘
                             ▼
                   ┌───────────────────┐
                   │  Foundry IQ       │  knowledge source
                   │  inv-knowledge    │  registered in the
                   │  (semantic mode)  │  inventoryiq project
                   └─────────┬─────────┘
                             ▼
                   ┌───────────────────┐
                   │ Foundry Agent     │  asst_v8Ye8v2iTPsBXRXWjDny8pRj
                   │ "InventoryIQ"     │  gpt-4o + instructions + grounding
                   │ (Foundry Agents)  │
                   └───────────────────┘

  ▼ Local fallback for judges without an Azure tenant
   ┌──────────────────────────────┐
   │  demo/  (Streamlit + DuckDB) │
   │  same schema, same questions │
   │  pastel themed UI            │
   └──────────────────────────────┘

Repo layout

Folder	What's in it
agent/	Declarative-agent manifest (manifest.json) + system-prompt instructions + action plugin definition. Used as documentation of the deployed agent and as a Copilot Studio import target if a Teams surface is added later.
grounding/scripts/	01_index_inventorymapper.py — ingests asset + alert rows into Azure AI Search with embeddings.
api/	(Reserved) .NET worker for real InventoryMapper → Search ingestion in production.
demo/	Local-runnable Streamlit chat over DuckDB. What reviewers without an Azure tenant use. Same schema, same questions, same answers.
docs/	Foundry setup guide + demo video script.

Microsoft IQ integration

This project integrates Foundry IQ as required by the hackathon rules.

Layer	Where it lives
Reasoning model	Azure AI Foundry project inventoryiq — gpt-4o deployment for chat, text-embedding-3-small for vectors
Knowledge source	Azure AI Search index inv-index registered as Foundry knowledge source inv-knowledge (semantic mode)
Grounding binding	Configured directly on the Foundry agent — every reply pulls grounded passages from inv-knowledge and cites them
Deployed agent	InventoryIQ (asst_v8Ye8v2iTPsBXRXWjDny8pRj) — Foundry-hosted, GA on the same Foundry project, testable in the built-in playground
Action skill (planned)	agent/actions/draft_procurement_ticket.json — turns a finding into a ticket draft

Why Foundry IQ over Fabric IQ: Fabric trial provisioning is currently gated for tenants created from Azure free signups (the trial dialog returned BadRequest for our region pair). Foundry IQ runs entirely inside Azure AI Foundry on the same Azure credit and provides the same grounding contract — a real RAG index with semantic ranking and source citations. The repo's grounding shape is identical either way, so a future Fabric IQ swap is a one-day addition.

Quick start — local demo (no Azure needed)

⚡ This local demo version does NOT require an Azure tenant, M365 subscription, or API keys. It is the primary way for judges to verify functionality if an Azure environment is unavailable.

cd demo
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
python seed.py            # builds inv.duckdb with ~500 assets, 30 days of monitoring, ~185 alerts
streamlit run app.py

Opens at http://localhost:8501.

Step-by-step demo testing

1. Setup environment — From demo/, install dependencies: pip install -r requirements.txt.
2. Generate mock data — Run python seed.py. Builds the local inv.duckdb populated with ~500 assets, 30 days of monitoring history, ~185 alerts.
3. Run the dashboard — streamlit run app.py launches the local chat interface.
4. Submit a query — Use the sidebar to click a sample question, or type a free-form one (e.g., "Which servers in Building B raised critical alerts?").
5. Observe grounding:
   • Cyan tool-call chips show which grounding function was invoked.
   • The data tables show the rows the agent grounded on.
   • The 📎 citations expander shows the lakehouse table + row IDs.
6. (Optional) Real LLM mode — Set AZURE_OPENAI_* in .env (see .env.example). With keys set, sidebar questions trigger real gpt-4o tool-calling with conversational synthesis instead of deterministic routing.
7. Theme toggle — Sidebar Theme button switches between dark and light pastel themes.

Quick start — Foundry IQ deployment

1. Provision Azure AI Foundry project + Azure AI Search Free + Azure OpenAI (see docs/foundry_setup.md). Total cost during the hackathon: < $5 of Azure free credit.
2. Configure .env from .env.example with your endpoints and keys.
3. Index the data:

   python grounding/scripts/01_index_inventorymapper.py --rebuild

   Reads from demo/inv.duckdb (or your real InventoryMapper SQL Server) and pushes ~685 documents into inv-index with embeddings via text-embedding-3-small.
4. Register inv-index as a Foundry knowledge source: in the Foundry project → Agentes → Create new → add Azure AI Search as a knowledge source → pick inv-index → name inv-knowledge → semantic search mode.
5. Configure the agent with the instructions in agent/instructions.md and bind the knowledge source.
6. Test in the Foundry Agents playground — see docs/demo_script.md for the suggested 3-minute demo flow.

Status

Track	Status
Local demo over DuckDB	✅ working
Mock lakehouse + seed (685 rows: 500 assets, 185 alerts, 30 locations, 9 blueprints)	✅ working
Grounding tools + citations (Streamlit demo)	✅ working
Streamlit chat with real gpt-4o tool-calling	✅ working
Pastel UI with dark + light theme toggle	✅ working
Declarative agent manifest (Copilot Studio compatible)	✅ drafted
Azure AI Search index inv-index	✅ deployed (685 docs indexed)
Foundry IQ knowledge source inv-knowledge	✅ connected (semantic mode)
Foundry-hosted agent InventoryIQ	✅ deployed + validated in playground
Demo video	🟡 in progress
README + repo polish + GitHub push	✅ deployed

Disclaimer

Built for demo purposes only. No confidential data. Schema, hostnames, users, alerts in this repo are synthetic. See hackathon disclaimer.

License

MIT — see LICENSE.