🚀 AI Foundry Enablement

Build fast. Iterate deeply. Scale meaningfully.

Quickstart • Architecture • Modules • Features • Roadmap

📅 This repository is actively maintained and updated frequently

Production-ready patterns for Azure AI Foundry with Semantic Kernel at the core: RAG, agentic systems, multimodal capabilities, and enterprise guardrails.

---

⚡ Quickstart

# Clone the repository
git clone https://github.com/<your-username>/ai-foundry-enablement.git
cd ai-foundry-enablement

# Set up Python environment
python -m venv .venv
source .venv/bin/activate      # Windows: .venv\Scripts\activate
pip install -r requirements.txt

# Configure environment variables
cp .env.template .env
# Edit .env with your Azure OpenAI or OpenAI credentials

# Run the Semantic Kernel Chat UI
python semantic_kernel/chat_ui.py

---

🏗️ Architecture

flowchart LR
    user[User / App] --> sk[Semantic Kernel]
    sk --> aoai[Azure OpenAI]
    sk --> search[AI Search]
    sk --> safety[Content Safety]
    sk --> agent[Agent Service]
    sk --> speech[Speech Services]
    sk --> obs[Observability]

Loading

---

📦 Modules

Module	Description
Semantic Kernel	Core orchestration with chat UI, plugins, and planners
RAG Pipeline	Baseline and agentic RAG with hybrid retrieval
Agent Service	Azure Agent Service integration
Multimodal	Speech-to-speech chat and video RAG capabilities
Observability	OpenTelemetry tracing and App Insights integration

---

✨ Features

• 🧠 Semantic Kernel Integration - Advanced orchestration with plugins and planners
• 🔍 Hybrid RAG - Vector + keyword search with semantic re-ranking
• 🎯 Agentic Workflows - Stepwise planning with tool orchestration
• 🔊 Multimodal Support - Speech and video processing pipelines
• 🔐 Enterprise Ready - Content safety, managed identity, and PII handling
• 📊 Full Observability - Distributed tracing and performance metrics

---

🧠 Semantic Kernel Patterns

Core Principles

1. Tools First - Single-purpose, stateless, idempotent functions
2. Strategic Planning - Choose the right planner for your use case
3. Structured Output - JSON schemas with Pydantic validation
4. Resilient Design - Timeouts, circuit breakers, and fallbacks
5. Streaming UX - Real-time token streaming with progress indicators

RAG Recipes

• R0 – Keyword - BM25 baseline, zero embeddings
• R1 – Vector - Cosine similarity on chunks
• R2 – Hybrid - BM25 + Vector with semantic re-ranking
• R3 – Agentic - Iterative retrieve → think → answer loops

---

📈 Performance

Scenario	Model	Latency	Accuracy
Chat	gpt-5	0.9s	-
Hybrid RAG	gpt-5-mini	1.6s	0.83
Agentic RAG	gpt-5	2.4s	0.88

---

🛡️ Security & Compliance

• Managed Identity for all Azure service calls
• Content Safety pre-screening on inputs and outputs
• Key Vault integration for secrets management
• PII Redaction before logging
• RBAC with minimal permissions

---

🗺️ Roadmap

• Vector database abstraction layer
• Advanced audio/video samples with benchmarks
• Evaluation dashboards and golden datasets
• Infrastructure as Code (Bicep) templates
• Multi-agent orchestration patterns
• LLM safety scorecards

---

📜 License

MIT © Ozgur Guler

---

🙏 Acknowledgements

Built on the Azure AI Foundry ecosystem and the Semantic Kernel community.

---

✨ Contact

For questions or contributions, reach out via LinkedIn or open an issue.