Machine Learning Engineer II · Microsoft
I build agentic AI systems and the large-scale data platforms they run on.
I'm Anshaj Khare, a Machine Learning Engineer II at Microsoft. I design and build production AI agents and cloud data platforms for supply-chain operations, combining retrieval and tool integration with multi-agent orchestration, evaluation, and safety guardrails.
16h → 1h
Recommendation latency after replatforming the circular cloud engine
400 hrs
Saved per developer per year by the data orchestration framework
100+
Microsoft engineers mentored on building scalable AI agents
4,000+
Professionals upskilled through teaching and guest lectures
About
Engineering reliable AI systems
I am a machine learning engineer working in Microsoft Cloud Supply Chain, where I design and ship agentic AI systems and the data platforms underneath them. My work spans MCP tool integration, retrieval, multi-agent orchestration, evaluations, and safety guardrails, alongside the backend and data engineering needed for reliable deployment.
Over the last few years that has meant replacing a 16-hour batch pipeline with a streaming platform that answers in under an hour, building an orchestration framework that removes hundreds of hours of pipeline plumbing per engineer per year, and putting real agentic systems in front of both internal services and end users.
Outside of work I write The Compute Blog, build side projects to scratch my own itches, and teach — around four thousand engineers and leaders so far, on agentic AI, generative AI, and machine learning.
Selected work
Things I've built
Side projects and open source where I get to make the architectural calls end to end.
The Compute Blog
A technical blog where I run experiments and write about how modern AI systems actually work — agent tooling, the Model Context Protocol, retrieval, and the engineering underneath them. Written for engineers who want the mechanism, not the marketing.
- Agentic AI
- MCP
- Systems writing
Streaks
A habit and learning tracker that makes deliberate practice stick. It pairs don't-break-the-chain streaks and daily goals with an SM-2 spaced-repetition scheduler, so solved problems and read topics resurface exactly when you are about to forget them. XP, levels, achievements and a year-long activity heatmap keep the loop going.
- Next.js 16
- TypeScript
- PostgreSQL
- Prisma 7
- Auth.js v5
- Tailwind CSS v4
- Zod
- Vitest
MCP Boilerplate
A reference implementation for the Model Context Protocol — a working starting point for wiring tools and resources into an MCP server.
- Python
- MCP
A2A Boilerplate
A demonstration of the Agent-to-Agent protocol, showing how independent agents discover each other and collaborate on a task.
- Python
- A2A protocol
RAG App
A template for building a scalable retrieval-augmented generation system, structured for production rather than for a demo notebook.
- Python
- RAG
- Vector search
Python Frameworks
A curated tour of powerful Python frameworks across web, data and ML, each with a runnable example — a map of the ecosystem for engineers deciding what to reach for.
- Python
Omdena — Student Debt
Worked as a machine learning engineer on Omdena's open collaborative platform, on a project studying the social and financial implications of student debt in the United States.
- Python
- Machine learning
Experience
Where I've done the work
Eight years across machine learning engineering, data platforms and cloud infrastructure.
Feb 2022 — Present
India
Current role
Machine Learning Engineer II
Promoted from L61 to L62Feb 2024
Building production AI agents and cloud data platforms for Microsoft Cloud Supply Chain, with end-to-end ownership spanning tool integration, retrieval, evaluations, and deployment.
MCP-enabled planning agents
Built production conversational agents that connect Model Context Protocol tools to Azure Data Explorer, helping teams explore operational data through natural-language questions.
Agentic RAG and human review
Developed retrieval-augmented agent workflows backed by vector databases to support operational decisions, combining automated outcomes with escalation to human review.
Multi-agent risk monitoring
Built production workflows with Microsoft Agent Framework to research external events, compare findings with tracked information, update retrieval stores, and surface operational risks or mitigations.
Evaluations and guardrails
Used Azure OpenAI and internal AI tooling for agent evaluations, safety guardrails, and production-readiness checks across deployed systems.
Platform modernization
Led the migration of the legacy engineering system that acts as the brain of the circular cloud operation, replacing SQL-based batch processing with Apache Spark, Kafka streams, a data lake and FastAPI service layers. The V-Next system produces intelligent recommendations in about an hour, down from sixteen.
Data orchestration framework
Built an event-driven framework that automatically creates, updates and deploys data pipelines and their supporting Azure resources — enabling high-scale enterprise ETL and saving roughly 400 hours per developer each year.
AI capability building
Mentored and upskilled 100+ engineers through targeted programs on scalable AI agents and modern agent frameworks. Received the org-level CSCP Stellar Quest award for end-to-end delivery of agentic systems.
ML platform for Lean Six Sigma
Architected the backend and MLOps layer for an in-house Lean Six Sigma product, covering model deployment and OpenAI integration.
- Python
- Apache Spark
- Kafka
- Azure
- Azure OpenAI
- Azure Data Explorer
- Microsoft Agent Framework
- MCP
- Agentic RAG
- Agent evaluations
- FastAPI
- Kubernetes
Dec 2019 — Feb 2022
India
Machine Learning Engineer
Shipped production ML services and cloud-native backends across fintech, SEO and e-commerce clients.
ML models in production
Built and deployed a transaction-classification service for a fintech product, containerized and served on AWS SageMaker, plus a spam-backlink detection model for an SEO business.
Serverless GraphQL API
Developed a complete serverless GraphQL backend deployed on Google Cloud Functions.
Distributed ETL
Implemented distributed ETL pipelines on Cloud Composer (managed Apache Airflow) and designed the relational schema for an e-commerce platform from the ground up.
Delivery engineering
Designed CI/CD pipelines for reliable deployment to GCP, and built resilient web scrapers for marketplace data collection.
- Python
- GraphQL
- AWS SageMaker
- Google Cloud
- Airflow
- Docker
Dec 2017 — Nov 2019
India
Data Engineer
Octathorpe Web Consultants
Modernized an analytics stack from spreadsheets to a cloud-native pipeline on Google Cloud.
Google Cloud migration
Migrated the analytics stack to GCP for reliability and consistency, with tighter BigQuery integration.
From Excel to Python
Moved analytics processes off Excel — first to R, then to Python — automating data extraction and powering visualization dashboards.
Event ETL
Built ETL pipelines to process user event data into relational stores backing the dashboards, along with Python mailers and notifiers that improved data visibility across the organization.
- Python
- R
- BigQuery
- Google Cloud
- SQL
Recognition
Named for the work, not the title
Two awards I'm proud of — one for putting AI agents into production, one for teaching.
Cloud Supply Chain Stellar Quest Award
AI Agents
Celebrating those who propel our Cloud Supply Chain into the AI-driven future by development of AI agents.
Presented by the Corporate Vice President for Cloud Supply Chain, for designing and shipping agentic systems that went into production use.
All Rounder Instructor
For his outstanding contributions in teaching, mentoring, and guiding learners with excellence, dedication, and impact.
Awarded by the founders for teaching across the full curriculum, from Python fundamentals through applied machine learning.
Writing
The Compute Blog
I run experiments and write up how modern AI systems actually work — agent tooling, evaluation, retrieval, and the engineering underneath. Explore a curated starting point and recent posts.
- Multi-Agent SystemsWhen to Let Your Agents Work as a Team
- Evaluating Agentic SystemsHow Do You Know Your Agent Actually Works?
- Gemini: Managed Agents & Interactions APIGoogle's newest solution to abstracting away the complexity of building agents
- Microsoft Agent FrameworkBuilding Production-Ready AI Agents in Python
Teaching & speaking
Explaining is part of the job
I've taught roughly 4,000 engineers and leaders — the fastest way I know to find the holes in your own understanding.
Technical Instructor
Interview Kickstart · Nov 2023 — Present
I teach working engineers preparing for senior roles at top technology companies, covering the full path from Python fundamentals through applied machine learning.
- Python
- Machine learning
- Deep learning
- Data analytics
Guest Lecturer & Mentor
Industry programs
I run technical programs for audiences ranging from individual software engineers to executive leadership across multiple organizations, translating fast-moving AI research into decisions teams can act on.
- Agentic AI
- Generative AI
- Machine learning
- Software engineering
AI Enablement
Inside Microsoft I designed and delivered the technical programs that took 100+ engineers from curiosity to shipping agentic systems, building the organizational readiness for AI-native product development.
- Agent frameworks
- Evaluation
- Production AI
Toolbox
What I reach for
Tools I've shipped production systems with, not just read about.
- Languages
- Python
- Go
- SQL
- TypeScript
- R
- AI & machine learning
- Agentic systems
- Multi-agent orchestration
- LLM applications
- Azure OpenAI
- Microsoft Agent Framework
- MCP
- RAG
- Agentic RAG
- Agent evaluations
- Safety guardrails
- Human-in-the-loop workflows
- MLOps
- Spark ML
- scikit-learn
- PyTorch
- Data platform
- Apache Spark
- Kafka
- Airflow
- Data lakes
- Streaming ETL
- Pipeline orchestration
- Cloud & infrastructure
- Azure
- Google Cloud
- AWS
- Kubernetes
- Docker
- Serverless
- CI/CD
- Unix
- Git
- Data stores
- Vector databases
- PostgreSQL
- MySQL
- SQL Server
- MongoDB
- BigQuery
- Azure Data Explorer
- ADLS
- HDFS
- SQLite
Certifications
Verified credentials
A sample of formal training in cloud, data engineering and machine learning.
AWS Certified Developer — Associate
Amazon Web Services
Developing, deploying and debugging cloud-based applications on AWS.
Data Engineering, Big Data and Machine Learning on GCP
Google Cloud · Coursera
Five-course specialization.
VerifyAdvanced Data Science with IBM
IBM · Coursera
Four-course specialization.
Verify