Build Your First AI Agent With Microsoft Azure AI Foundry

The trainer. Microsoft Certified Trainer (MCT) and AI/ML technical trainer at Koenig Solutions, accredited to deliver official Microsoft AI, data, and cloud curricula and rated a consistent 5/5 by learners. 750+ professionals trained globally — Singapore Ministry of Defence, Bank of Uganda, the Federal Reserve Bank of Atlanta, Microsoft's Unnati 2.0 AI upskilling program, TCS, HCL Tech, Wipro, DXC Technology, Orange Business, Nestlé, KPMG, EY, and Deloitte. Certified across the AI track (AI-103, AI-901), Azure engineering (AZ-204, AZ-104, AZ-900), and data, security, and analytics (DP-900, SC-900, PL-300), with a B.S. in Data Science and Applications from IIT Madras.

What I teach. AI method selection and systems design — when classification, extraction, retrieval, generation, or an agent is the right tool, and when a rules engine beats all of them. LLM application development with retrieval grounding and tool use; production RAG with embeddings and pgvector similarity search; agentic systems, tool calling, and MCP server design; prompt engineering and output evaluation for hallucination risk and compliance; the Claude developer stack and the full Microsoft Copilot stack; and the Microsoft data platform via Azure AI Foundry, Fabric, Databricks, and Power BI.

I build, and I teach. The training is grounded in shipped software, not slideware. A practitioner of AI-assisted structural development, I use Claude Code as my primary development partner to architect, iterate, and ship production-grade full-stack applications — the developer stays the architect and makes every design decision; the AI collapses weeks into days. Six applications shipped on that model, so the labs come from real trade-offs: why a RAG pipeline falls back to a live crawl instead of re-embedding, why a memory cap sits in front of a bulk data pull. Enterprise teams get both halves of the job — the build, and the judgment about governance, cost, and when not to use AI at all.

TRAINING DELIVERY SNAPSHOT

• 750+ professionals trained across the globe through instructor-led, hands-on technical enablement.
• 5/5 average learner rating, cited for depth of topic knowledge, patience, and adapting content live to team needs.
• Government & defence: Singapore Ministry of Defence · Bank of Uganda · Federal Reserve Bank of Atlanta
• Microsoft programs: Unnati 2.0 — Microsoft AI upskilling program.
• Consulting & Big Four: EY · KPMG · Deloitte
• IT services & enterprise: TCS · HCL Tech · Wipro · DXC Technology · Orange Business · Nestlé
• Delivery model: remote and global, across time zones, for engineering and strategy audiences.

AI/ML TRAINING DELIVERY AREAS

AI Method Selection & Systems Design Classification, extraction, summarization, retrieval, generation, and agentic orchestration; matching AI versus rules-based approaches to real business use cases.

LLM Application Development LLM-powered summarization, extraction, Q&A, and drafting features; retrieval grounding, tool use, and integration into FastAPI/Flask backends.

RAG, Embeddings & Vector Databases Embedding generation, pgvector similarity search, retrieval pipelines with live-crawl fallback, and index/vector-store trade-offs for production RAG.

Agentic AI, Tool Calling & MCP Tool calling, building and connecting MCP servers, multi-agent orchestration, human-in-the-loop approval and fallback design, and agentic RAG.

Claude Developer Stack Claude Code for agentic software development; MCP server design and connectivity; Claude Projects, Skills, and Artifacts for reusable, context-grounded workflows.

Microsoft Copilot Stack GitHub Copilot (IDE and agent mode); Microsoft 365 Copilot adoption and productivity patterns; Copilot Studio for custom agents; Azure AI Foundry and Copilot extensibility.

AI-Assisted Structural Development Using Claude Code as a development partner on production codebases — architecture-first workflows, scoping and reviewing AI-generated changes, keeping the developer as system architect, and the failure modes teams hit when they invert that relationship.

Prompt Engineering Prompt template design, iteration, and evaluation for tone/format/CTA; reviewing outputs for compliance, hallucination risk, and brand alignment.

Data Platform, BI & Analytics Microsoft Fabric, Databricks, and Power BI enablement — data engineering workflows, semantic modelling, and analytics delivery for enterprise teams.

ML Lifecycle & Dataset Preparation Training–validation–deployment flow, feature/label alignment, data lineage and readiness, and retraining-trigger thinking for managed ML systems.

Scalable & Real-Time Data Async and event-driven pipeline design, caching/TTL strategy, memory-safe high-volume ingestion, and streaming-versus-micro-batch trade-offs.

Applied AI on Azure & Cloud DevOps Azure AI services and AI Foundry, App Service and Blob Storage, containerization with Docker, and CI/CD with Azure DevOps for shipping AI-enabled applications.


 

APPLIED AI/ML PROJECTS

StockiNews — NSE Portfolio & AI-Powered Financial News (RAG)

React 19 · FastAPI · Supabase · pgvector · RAG · yfinance · Chart.js

• Built a RAG pipeline that embeds news articles into a Supabase pgvector store; queries retrieve semantically relevant articles first and fall back to live web crawlers only for fresh content, significantly reducing token usage.
• Delivered 25+ FastAPI endpoints over Supabase (PostgreSQL + pgvector) with 47 smoke tests and a memory-safe pipeline hard-capped to prevent OOM on large data pulls.
• Tracked NSE portfolios with correct cost-basis P&L across BUY/SELL transactions; integrated live pricing via yfinance with a per-endpoint file cache (6h–48h TTLs).

PharmaFlow — Multi-Tenant Pharmacy Operations Platform

React 18 · Flask · PostgreSQL · Google Cloud Vision OCR · SQLAlchemy · Docker

• Implemented zero-entry invoice scanning with Google Cloud Vision OCR, extracting line items from distributor photos and PDFs using expiry date as a parsing anchor — a production document-extraction AI use case.
• Built FEFO inventory allocation (soonest-expiring batches first) aligned to Indian pharmacy regulations, with auto-generated printable debit notes and compliant invoices.
• Architected multi-tenancy with three-tier RBAC (Owner/Manager/Staff) and an analytics dashboard breaking down revenue, COGS, and profit across 7/30/90/365-day windows.

PLATFORM & PRODUCT ENGINEERING PROJECTS

KiteKonnect — Internal Knowledge & Community Platform (Koenig Solutions)

React 19 · Node.js · Socket.IO · Tailwind CSS v4 · Azure Blob Storage

• Built