Enterprise-grade AI services powered by Azure Cognitive Services.

🔗 Live Demo: https://azure-ai.streamlit.app

┌──────────────────────────────────────────────────────┐
│                   Streamlit UI                       │
├──────────────────────────────────────────────────────┤
│               Service Abstraction                    │
│  ┌────────────────────────────────────────────────┐  │
│  │              AzureBaseClient                   │  │
│  │  • Retry logic (429, 5xx)                      │  │
│  │  • Timeout handling                            │  │
│  │  • Request tracing (x-client-request-id)       │  │
│  └──────────┬──────────────────┬──────────────────┘  │
│             │                  │                     │
│    ┌────────┴──────┐  ┌────────┴──────┐              │
│    │LanguageClient │  │ VisionClient  │              │
│    └───────────────┘  └───────────────┘              │
│    ┌───────────────┐  ┌───────────────┐              │
│    │ SpeechClient  │  │DocIntelClient │              │
│    └───────────────┘  └───────────────┘              │
├──────────────────────────────────────────────────────┤
│  Telemetry (AzureCallMeta)                           │
│  • Service, operation, latency, status, request ID   │
└──────────────────────────────────────────────────────┘

• Service Abstraction: Shared base client with retry/timeout logic; per-service subclasses
• Structured Telemetry: Every API call returns AzureCallMeta (request ID, latency, status)
• Resilience: Automatic retry on transient failures (429, 500–504)
• Observability: Sidebar diagnostics panel with call history

Copy the example secrets file and fill in your Azure credentials:

cp .streamlit/secrets.toml.example .streamlit/secrets.toml

git clone https://github.com/tvprasad/azure-ai-studio.git
cd azure-ai-studio
pip install -r requirements.txt

cp .streamlit/secrets.toml.example .streamlit/secrets.toml
# Edit with your Azure credentials

streamlit run app.py

• Added Speech Services: fast Speech-to-Text transcription and neural Text-to-Speech synthesis
• Added Document Intelligence: prebuilt-read model for PDF and image extraction
• Promoted AzureBaseClient shared retry/timeout/telemetry base class
• Updated sidebar diagnostics with total call counter
• Updated architecture and configuration documentation

• Added AI Governance layer: GovernanceLayer class with per-request cost estimation and session-level usage aggregation
• Added Governance sidebar panel: real-time usage metrics showing total calls, session cost, and per-service breakdown
• Expanded AzureCallMeta with estimated_cost_usd field populated on every API call
• Added _GOVERNANCE_PRICING table (Azure retail pricing, 2026-Q1) covering all four services
• Updated AzureSpeechClient.fast_transcribe and synthesize to return (result, AzureCallMeta) tuples
• Enhanced diagnostics panel to include estimated cost per call
• Cost captions added to all service result views
• Billing unit semantics: Language=text record, Vision=transaction, Speech STT=minute, Speech TTS=character, Document Intelligence=page

• Added Vision Intelligence: image analysis, tagging, OCR via Azure AI Vision
• Added AzureBaseClient with shared retry and telemetry logic
• Introduced AzureVisionClient

• Enterprise service layer: AzureLanguageClient with retry, timeout, and request tracing
• Structured telemetry: AzureCallMeta dataclass
• Sidebar diagnostics panel
• Caching via st.cache_data

• Initial release: Language Intelligence (Sentiment, Entities, Key Phrases, Language Detection)

Prasad Thiriveedi

20+ years in distributed event-driven systems. Now leading teams and architecting enterprise AI infrastructure.

GitHub | LinkedIn

MIT