21 min read
Fabric Data Agent Accuracy Is a Distribution, Not a Number
I asked a Fabric data agent 24 questions in three wordings, ten times each, on real AEMO data. It scored 82%, but only 39% of wordings were always right.
Python
119 articles
8 min read
Fabric Data Agents Move to the Responses API: What Consumers Must Change
Fabric data agents now use GPT-5.1 and a Responses API client. What notebook, Foundry and app callers must change, and why you need a full eval re-run.
8 min read
Harness Agent or Plain Agent? What Agent Framework Now Decides for You
The Agent Framework Harness bundles planning, memory, compaction, approvals and telemetry. When to accept its opinions and when a plain Agent is better.
7 min read
Agent Fallbacks Are a Launch Requirement, Not an Ops Task
Decide what an agent does when a tool fails, a step runs out of budget or grounding is weak before launch, and make the release check fail without it.
7 min read
Output Contracts: Catch a Model Swap Before Downstream Systems Do
Structured outputs fix the shape of a model's answer, not its meaning: a two-layer contract between a Foundry model and the systems that consume it.
8 min read
Where Agent State Breaks: Resume, Retries and Duplicate Runs
Typed state and checkpoints don't stop an agent paying twice: five places agent workflow state breaks at its boundaries, and patterns that hold.
9 min read
RAG Citations People Can Check: Treat Them as Part of the Answer
Good retrieval is wasted if users cannot verify the answer, so design RAG citations to be checkable, validated in code and measured like relevance.
8 min read
Turning Production Queries into an LLM Eval Set You Can Trust
How I sample real user queries by stratum, redact PII before anyone reads them, split off a holdout and keep the eval set current as the product changes.
8 min read
Canary a Model Swap in Foundry: Sticky Splits and Version Stamps
An offline evaluation can't tell you how a new model handles real traffic. How I'd canary a Foundry model change with sticky splits and per-call version logs.
8 min read
Hybrid Search for RAG: Test What the BM25 Leg Rescues
Hybrid search is default RAG advice, but the BM25 leg has real costs; a three-way test on Azure AI Search shows which questions it rescues.
8 min read
Testing Agent Tool Policies: Prove the Limits Hold
A tool policy nobody tests is a comment. How to assert agent tool lists and approvals in CI, then gate tool use on a golden set of scenarios.
8 min read
Treat Retrieval Changes Like Schema Changes: A RAG Regression Gate
Make retrieval-first a team rule: an owned question set, a side-by-side index comparison on Azure AI Search, and a gate that fails on lost answers.
9 min read
Model Retirements in Foundry: Choose the Upgrade Policy on Purpose
Every Foundry model deployment has a version upgrade policy, and the default swaps your model at retirement. How to choose, audit and run model changes safely.
6 min read
Explicit State for Agent Workflows: Stop Making the Model Remember
Agents get confused when workflow state lives in the chat transcript. Four patterns for typed state, fresh context and code-owned transitions.
8 min read
Is Your Reranker Earning Its Place? Test It Before You Pay for It
The semantic ranker only reorders the top 50 results. A with-and-without test on a labelled set shows which RAG queries it rescues and which it just slows.
6 min read
Least-Privilege Tools for AI Agents: Four Layers of Control
Agent reliability depends on which tools a model can reach. Four layers of tool policy (surface, schema, approval, budget) with Agent Framework code.
7 min read
Wrong RAG Answer? Rule Out Retrieval Before You Edit the Prompt
Most wrong RAG answers start with the passages, not the prompt. A labelled question set and a retrieval-only triage on Azure AI Search show which one to fix.
6 min read
Workflow First, Agent Inside: Putting the Model in One Step
Most agent designs only need autonomy in one step. How to build the workflow around it with Agent Framework executors, switch-case edges and loop caps.
8 min read
Agentic Retrieval or a Hand-Built RAG Pipeline? How I Decide
Azure AI Search knowledge bases now plan queries for you, in preview. When to hand over the RAG pipeline, when to keep building it, and how to test both.
7 min read
Boring Code on Purpose: Writing Software I Can Maintain at 2 AM
Why I now prefer explicit, predictable code over clever one-liners, the four questions I ask before merging, and when terse code is the better choice.
9 min read
Seven RAG Failure Modes and the Pattern That Fixes Each
Diagnose why a RAG system fails before adding patterns: hybrid search, semantic reranking, grounding gates and citations on Azure AI Search, mapped to failures.
1 min read
Building a Personal AI Assistant with Semantic Kernel
This foundation can be extended with calendar integration, email access, and more plugins. Perfect for a holiday coding project!
1 min read
Migrating from LangChain to Semantic Kernel: A Practical Guide
Many teams that started with LangChain in 2024 are now evaluating Semantic Kernel for its tighter Azure integration and production-ready features. Having…
1 min read
Testing LLM Applications: Strategies Beyond Traditional Unit Tests
LLM outputs vary between runs and model versions. Testing must focus on behavioral properties rather than exact string matching, while still catching…
1 min read
Azure Functions with AI: Building Serverless AI Applications
Configure appropriate timeout settings for AI operations and implement retry logic for transient failures. Consider using premium plans for…
1 min read
Azure Document Intelligence: Table Extraction and Analysis
Table extraction transforms static reports into queryable data. Combine extracted tables with LLM analysis to answer questions about financial statements…
1 min read
Azure AI Document Intelligence: Extracting Structured Data from Invoices
First, provision an Azure AI Document Intelligence resource and configure your client. The prebuilt invoice model recognizes vendor details, line items…
1 min read
Structured Output with Azure OpenAI: JSON Mode and Response Formats
JSON mode guarantees the model outputs valid JSON, though you still need to specify the schema in your prompt.
1 min read
Function Calling Patterns in Azure OpenAI
Functions are defined as JSON schemas that describe parameters and their types. Well-designed function schemas and clear descriptions ensure the model calls…
1 min read
Building Production MLOps Pipelines with Azure Machine Learning
Automated pipelines ensure reproducibility, enable A/B testing, and provide audit trails for model governance.
1 min read
Building RAG Applications with Azure AI Search and GPT-4o
RAG works by first retrieving relevant documents from a search index, then passing those documents as context to an LLM for generation. This approach…
1 min read
Building AI-Powered Search with Azure AI Search and Semantic Ranker
Monitor reranker scores and caption extraction quality. A/B test semantic ranking against pure vector or keyword search to quantify improvements for your…
1 min read
Multimodal RAG: Combining Text, Images, and Tables for Enhanced Search
Multimodal RAG ensures users find relevant information regardless of how it's represented in the source documents.
1 min read
Structured Output with JSON Mode: Reliable Data Extraction from LLMs
Structured outputs eliminate parsing errors. Add business validation for semantic correctness - the schema ensures format, you ensure meaning.
1 min read
Implementing Semantic Caching for LLM Applications
Use Redis Vector Search for efficient similarity matching at scale. Tune the similarity threshold based on your use case - higher values ensure more precise…
1 min read
Building Knowledge Graphs with Azure OpenAI and Neo4j
Graph RAG significantly improves answers for questions involving relationships, hierarchies, and multi-hop reasoning.
1 min read
Azure Functions with AI: Serverless Inference Patterns
Use consumption-based billing to pay only for actual inference time. Implement caching for repeated queries and batch similar requests to amortize cold…
1 min read
Claude Extended Thinking: Solving Complex Reasoning Problems
Enable extended thinking for mathematical proofs, code analysis, strategic planning, and multi-constraint optimization problems. For simple queries…
1 min read
Azure Machine Learning Managed Endpoints: Deploying Models with Zero Downtime
Configure Azure Monitor alerts on latency and error rates. Implement automatic rollback triggers when thresholds are breached.
1 min read
Implementing Guardrails for Production LLM Applications
Guardrails are essential for enterprise AI deployments, protecting both users and your organization.
1 min read
Fine-Tuning GPT-4o for Domain-Specific Tasks on Azure
After training completes, evaluate on a held-out test set before deploying. Monitor the fine-tuned model's performance against the base model to ensure…
1 min read
Evaluating RAG Systems: Metrics and Automated Testing with Azure AI
Fail builds when quality metrics drop below thresholds, catching regressions before they reach production.
1 min read
Building Multi-Agent Systems with Azure AI Foundry Orchestration
Azure AI Foundry provides trace visualization showing the complete agent interaction graph, making it easy to debug complex workflows and optimize handoff…
1 min read
Prompt Caching with Claude and Azure OpenAI: Reducing Latency and Costs
Track cache hit rates in your monitoring. Applications with high context reuse typically see 60-80% cache hit rates, dramatically reducing per-request costs.
1 min read
Cosmos DB Vector Search: Implementing Semantic Search at Global Scale
Configure your Cosmos account with write regions near your users and read replicas globally. Vector searches automatically route to the nearest replica…
1 min read
Real-Time AI Inference with Azure Container Apps: Dynamic Scaling Patterns
Use spot instances for non-critical workloads, implement request batching for throughput optimization, and configure appropriate scale-to-zero policies.…
1 min read
Azure AI Search Vector Optimization: Reducing Costs While Improving Recall
Vector search costs can spiral quickly at scale. After optimizing Azure AI Search deployments processing 50 million vectors, I've identified key patterns…
1 min read
GPT-4o Vision for Document Intelligence: Building Smart OCR Pipelines
For multi-page documents, process pages in parallel and use a synthesis step to merge extracted data. GPT-4o handles cross-page references like "continued…
1 min read
Claude 3.5 Sonnet for Code Review: Automated PR Analysis Patterns
Claude 3.5 Sonnet has become my go-to model for automated code review workflows. Its exceptional ability to understand context across large codebases makes…
1 min read
Building AI Agents with Azure AI Foundry: A Post-Build 2025 Implementation Guide
The new agent framework introduces several critical improvements. First, native multi-agent orchestration allows agents to collaborate without custom…
7 min read
Turning Call Recordings into Data with Azure AI Speech and GPT-4o
A batch pipeline for call and meeting audio: Azure AI Speech transcription with diarization, then GPT-4o structured outputs into rows you can query.
7 min read
LLM-Drafted Data Dictionaries: Profile First, Describe Second
Draft column descriptions for an Azure SQL database with Azure OpenAI, grounded in profiling stats, reviewed by a data owner, then stored as MS_Description.
8 min read
LLM-Drafted Tests for PySpark: The Model Writes, pytest Judges
Use Azure OpenAI structured outputs to draft edge-case test data for PySpark transforms, review it like code, and let pytest and assertDataFrameEqual decide.
7 min read
LLM-Drafted Data Quality Rules: Let the Model Propose, Not Enforce
Use Azure OpenAI structured outputs to draft Great Expectations rules from a data profile, then backtest and review them before they ever gate a pipeline.
6 min read
Azure AI Foundry from Python: A Data Team's First Project
A practical first project in Azure AI Foundry for data teams: hubs and projects, the preview azure-ai-projects SDK, and an evaluation loop from day one.
7 min read
Model Context Protocol Six Weeks In: What It Is and Where It Fits
A practitioner's read on Anthropic's Model Context Protocol in January 2025: the primitives, a read-only SQL server in Python, and when not to use it.
1 min read
Error Handling in LLM Applications: A Comprehensive Guide
Robust error handling is what separates prototypes from production systems. Invest in comprehensive error handling early to avoid painful debugging later.
1 min read
Reliable Data Extraction with LLMs: Patterns and Practices
Reliable extraction requires careful schema design, multi-pass validation, and explicit handling of uncertainty. These patterns help you build extraction…
1 min read
Building Type-Safe AI Applications: From Prompts to Production
Type safety transforms AI applications from fragile prototypes into robust production systems. Invest in types early - your future self will thank you.
1 min read
Pydantic Integration with OpenAI: Type-Safe AI Development
Pydantic integration makes OpenAI's structured outputs truly type-safe, catching errors at development time and ensuring your AI-generated data is always valid.
1 min read
Conditional Edges in LangGraph: Dynamic Agent Flows
Route based on multiple state attributes: Sometimes you don't know all routes ahead of time: Multiple decision points in sequence: Route with randomization…
1 min read
LangGraph Introduction: Building Stateful AI Agents
LangGraph provides these capabilities through a graph-based execution model.
1 min read
LangChain Updates: What's New in July 2024
LangChain 0.2.x Stability LCEL (LangChain Expression Language) Maturity Azure AI Search Vector Store
1 min read
Fabric SDK: Python Development for Microsoft Fabric
The Microsoft Fabric SDK provides a Pythonic interface for working with Fabric services. This guide covers installation, configuration, and common…
8 min read
Azure AI Search Index Operations: Schema Changes, Rebuilds and Aliases
Which Azure AI Search index changes work in place, which force a rebuild, and how to swap versioned indexes, load documents and check health safely.
7 min read
Azure AI Search Semantic Ranker: Configuration, Thresholds and Cost
How to design semantic configurations, use reranker scores as a RAG cut-off, and decide when Azure AI Search's semantic ranker is worth paying for.
6 min read
CrewAI 0.1 on Azure OpenAI: Roles, Tasks and Current Limits
A practical look at CrewAI 0.1.x with Azure OpenAI: how agents, tasks and sequential crews work, what's missing today, and when to choose it.
7 min read
AutoGen 0.2 on Azure OpenAI: Termination, Code Execution and Tools
A practical look at AutoGen 0.2 with Azure OpenAI: wiring the config, stopping runaway conversations, sandboxing generated code and registering tools.
6 min read
Self-Correcting RAG: Grade Retrieval Before You Answer
How to add retrieval grading, query rewriting, and groundedness checks to RAG on Azure OpenAI and Azure AI Search without tripling latency or cost.
6 min read
Structure-Aware Chunking for RAG: Headings, Tables and Breadcrumbs
Chunk documents along their headings, keep tables whole and prefix every chunk with its section path, using Document Intelligence markdown output and Python.
7 min read
Owning the Loop: AI Assistant Architecture on Chat Completions
How to structure a production AI assistant on Azure OpenAI Chat Completions: conversation state, retrieval, tool calling and the orchestration loop you own.
8 min read
Assistants API Runs in Production: Polling, Tool Calls and Expiry
How to drive OpenAI Assistants API runs reliably in January 2024: the run state machine, polling, function calls, expiry, concurrency and cost.
7 min read
Shipping on GPT-4 Turbo Preview in Azure OpenAI: A Production Checklist
GPT-4 Turbo is still a preview model on Azure OpenAI; here's how I'd deploy, budget, retry and test around 1106-preview before real users touch it.
1 min read
Code Interpreter: AI-Powered Data Analysis
Code Interpreter (Advanced Data Analysis) is the single most productive AI tool I've used for exploratory analysis — it runs Python in a sandbox, opens…
1 min read
Understanding Runnable Interfaces in LangChain
Deep dive into LangChain's Runnable interface and its implementations for building flexible LLM pipelines.
1 min read
LangChain Expression Language (LCEL): A Practical Guide
Understanding LangChain Expression Language for building composable LLM applications with clean, declarative syntax.
1 min read
SemPy Library: Python for Semantic Models in Fabric
Deep dive into the SemPy library for working with Power BI semantic models in Microsoft Fabric notebooks.
1 min read
Data Science Workflows in Microsoft Fabric
Building end-to-end data science workflows using Microsoft Fabric's integrated capabilities.
1 min read
LangChain with Azure OpenAI: Building Production AI Applications
Learn how to use LangChain with Azure OpenAI Service to build robust, production-ready AI applications.
1 min read
LangChain Updates: New Features and Improvements
LangChain's velocity in 2023 has been remarkable and occasionally destabilising — the 0.0.x series has introduced breaking API changes in point releases…
1 min read
PEFT Libraries: Practical Tools for Efficient Fine-Tuning
Tomorrow we'll explore Hugging Face integration with Azure. PEFT Documentation PEFT GitHub TRL Library
1 min read
Building AI Agents: From Chatbots to Autonomous Systems
AI agents represent the next evolution of AI systems. By combining planning, tool use, and memory, they can autonomously accomplish complex goals while…
1 min read
AI Orchestration Patterns: Coordinating Multi-Model Systems
AI orchestration patterns enable building sophisticated AI systems from modular components. The key is designing for reliability, observability, and…
1 min read
AI-Powered Data Quality: Intelligent Data Validation
AI-powered data quality goes beyond static rules. Intelligent systems understand context, detect subtle anomalies, and provide actionable recommendations…
1 min read
Spark ML Patterns for Production Systems
Spark ML provides battle-tested patterns for production machine learning. From feature engineering to model persistence, these patterns ensure reliable ML…
3 min read
AI Assistants in Jupyter and Databricks Notebooks
Error: {type(error).name}: {str(error)} response = openai.ChatCompletion.create( engine="gpt-4", messages=[{"role": "user", "content": prompt}] ) return…
1 min read
Building AI Data Analysis Assistants
AI data analysis assistants make insights accessible to everyone, regardless of technical expertise. The key is combining natural language understanding…
1 min read
LangChain with Azure OpenAI: Getting Started
Combined with Azure OpenAI's enterprise security and compliance, it's a powerful combination.
1 min read
LangChain Introduction: Building AI Applications with Python
Create reusable, parameterized prompts: Combine components into workflows: Load content from various sources: Split documents for embedding: Generate…
1 min read
TDD (Test-Driven Development) Overview with Python Example
This cycle is repeated for each new piece of functionality. The result is a suite of tests that provide confidence in the code and make it easier to…
7 min read
RAG on Azure OpenAI: Grounding GPT-3 Answers in Your Own Documents
What retrieval augmented generation is, why it beats fine-tuning for company knowledge, and a minimal loop on Azure OpenAI completions in January 2023.
6 min read
Qdrant 0.11 with Azure OpenAI: Point IDs, Payload Indexes and Filters
Run Qdrant 0.11 with Azure OpenAI embeddings: valid point IDs, payload indexes, filtered HNSW search, memory settings, and when Qdrant isn't the right choice.
7 min read
Milvus 2.2 with Azure OpenAI Embeddings: A Self-Hosted Start
Run Milvus 2.2 standalone, load Azure OpenAI embeddings with pymilvus, and learn the schema, index, consistency and delete rules before you self-host it.
7 min read
Weaviate 1.17 with Azure OpenAI: Bring Your Own Vectors
Run Weaviate 1.17 with embeddings from Azure OpenAI: schema choices, batch imports, filtered vector search, BM25 and the hybrid search caveats to know first.
8 min read
Pinecone With Azure OpenAI Embeddings: Design the Index First
Set up a Pinecone index for Azure OpenAI embeddings: dimensions, pod types, namespaces, metadata indexing and idempotent upserts, plus when to skip it.
8 min read
Semantic Search Prototype: Azure OpenAI Embeddings and NumPy
Build a small semantic search prototype on Azure OpenAI embeddings and NumPy, measure it with hit rate@k, add BM25, and know when to move to a real index.
8 min read
Text Embeddings on Azure OpenAI: What They Are and When to Use Them
What text embeddings are, which Azure OpenAI embeddings models you can use in January 2023, how to compare vectors, and when embeddings are the wrong tool.
7 min read
Streaming Azure OpenAI Completions from the API to the Browser
How stream=True works on the Azure OpenAI 2022-12-01 Completions API, what you give up by streaming, and how to relay tokens to a browser without losing any.
7 min read
Calling Azure OpenAI over Plain HTTP: The 2022-12-01 REST Contract
What the GA 2022-12-01 Azure OpenAI REST API looks like on the wire: URLs, auth headers, completions, embeddings, errors and when to skip the SDK.
7 min read
Azure OpenAI Is GA: What Changed on 16 January and What Didn't
Azure OpenAI Service reached GA on 16 January 2023: what that changes for production plans, what still needs an approved application, and how to proceed.
7 min read
Azure OpenAI from Python: Setting Up the openai 0.26 Library Properly
How to configure the openai 0.26 Python library for Azure OpenAI in January 2023: API keys or Azure AD tokens, deployment names, timeouts, retries and async.
6 min read
Completions Now, Chat Later: Keeping Azure OpenAI Code Portable
Azure OpenAI is GA with a Completions API only, and ChatGPT is coming soon. How to decide what needs chat and keep your prompt code ready for the change.
6 min read
Temperature and Top-P in Azure OpenAI: What They Actually Change
How temperature and top_p reshape token sampling in Azure OpenAI completions, why to tune one not both, and how to measure the effect before you ship.
8 min read
No System Prompt Yet: Instruction Preambles on Azure OpenAI
Azure OpenAI completion models have no system role yet. How to design, budget, version and test an instruction preamble that does that job instead.
8 min read
Few-Shot Prompts on Azure OpenAI: Choosing and Ordering Examples
How to pick, order and budget few-shot examples for Azure OpenAI completion models, with the biases to watch for and when to stop adding examples.
8 min read
Counting and Capping Tokens in the Azure OpenAI Preview
How to count tokens with tiktoken, size max_tokens, log real usage and cap spend for Azure OpenAI completion models in the January 2023 preview.
7 min read
Throttling in the Azure OpenAI Preview: Pacing Per-Deployment Requests
Azure OpenAI preview limits are requests per second per deployment. How to pace calls, retry 429s properly and scale out with the openai 0.26 Python library.
7 min read
Handling Content Filtering in the Azure OpenAI Completions API
How Azure OpenAI's preview content filter shows up in code: HTTP 400 prompts, content_filter finish reasons, and why you still need your own guardrails.
8 min read
Responsible AI on Azure OpenAI: What the Platform Covers and What You Own
Azure OpenAI's preview ships with use-case review, content filtering and abuse monitoring. Here is where those controls stop and your application's begin.
7 min read
Azure OpenAI vs OpenAI API: What Differs in January 2023
Azure OpenAI or OpenAI's own API in January 2023? Access, models, data use, cost and code compared, plus one Python client that can target either.
7 min read
ChatGPT-Style Chat on Azure OpenAI Without a Chat API
There is no ChatGPT API yet. Here is how to build multi-turn chat on Azure OpenAI's Completions API: transcripts, state, token budgets and streaming.
8 min read
GPT-3.5 on Azure OpenAI: What You Get Before text-davinci-003
What GPT-3.5 means on Azure OpenAI in January 2023: which models are in the catalogue, what they cost, and how to design for text-davinci-003's arrival.
7 min read
GitHub Secret Scanning: What to Do When the Alert Fires
How GitHub secret scanning works in early 2022, what it misses, and a practical playbook for triaging alerts, custom patterns, and rotating leaked keys.
1 min read
Databricks REST API for Advanced Automation
The Databricks REST API exposes every workspace capability that the UI and CLI provide—and then some—making it the integration point for external…
9 min read
Using WSL2 (Windows Subsystem for Linux) for Blockchain Development
In the past, I've been using my Macbook Pro to do all things blockchain and smart contracts development. One of the hurdles I encountered in the past is…
1 min read
PySpark Best Practices for Production Workloads
1. Define schemas explicitly Never rely on schema inference in production 2. Use built in functions Avoid UDFs when possible 3. Write pure transformation…
7 min read
Triaging Azure Advisor Cost Recommendations Before You Act
Which Azure Advisor cost recommendations are safe to act on, which need judgement, and how to pull, tune and suppress them with the CLI and Python SDK.
8 min read
Form Recognizer Invoices in Production: Confidence Gates and Review
How to put Form Recognizer's preview invoice model behind confidence thresholds, business-rule checks and a human review queue before data reaches the ERP.
1 min read
Azure Functions with Python: Getting Started
.NET on Functions is the default story Microsoft tells, but in practice the serverless workload that lands on my desk most often is a Python data scientist…