Hi, I'm Varun

 

I build agentic AI systems: planners, tool routers, memory, and the infrastructure that ships them to production.

years shipping
3+

years shipping

Best Engineer, MSTR
2×

Best Engineer, MSTR

B.E. CGPA
9.57

B.E. CGPA

View work Résumé

Who's building this.

operator_profile.json

class
AI Engineer
specialty
Agentic workflows, RAG, MCP, evals
current_focus
Agent stack at MicroStrategy
education
B.E. IT, VESIT (2020–2024)
email
varun.motiyani14@gmail.com

Software engineer with 3+ years building scalable backends and high-performance APIs in Python, Django, and FastAPI, now focused on applied GenAI: agentic workflows, RAG pipelines, MCP integrations, and evaluation frameworks. Currently AI Engineer at MicroStrategy, where I architect the agent stack behind AutoAgents and AutoDash.

Before that: Associate Software Engineer at Saint-Gobain INDEC, then Full-Stack Developer Intern at Edelweiss Housing Finance.

Projects

View work →

Where I've worked

Full timeline →
MicroStrategy Inc.Jul 2025 to Present  /  IDC Pune

AI Engineer 1

50%+lower agent latency
  • ▸Sole architect of a ReWOO + fallback ReACT agentic architecture: a tool-decoupled plan/execute pipeline (web, geospatial, proximity-search modules) that resolves complex interdependent queries, cutting response latency 50%+ while improving accuracy.
  • ▸Built end-to-end Model Context Protocol integration with fastmcp, turning AutoAgents into a compliant MCP client/host so users can connect Atlassian, GitHub, Notion, Mosaic and other MCP servers.
  • ▸Designed SimpleMem, a lightweight multi-tenant agent-memory layer (short and long term) on ClickHouse and Redis, shipped as a reusable package adopted across AutoAgents, AutoDash and CustomerExpert.

  • Associate Software Engineer

    Saint-Gobain INDEC  /  Jun 2024 to Jul 2025

    70% faster query resolution

  • Full-Stack Developer Intern

    Edelweiss Housing Finance  /  Dec 2022 to May 2023

    +15% extraction accuracy

Things I keep coming back to.

Working notes drawn from the systems, constraints, and production questions behind the work above.

Orchestration

The handoff is where an agent system earns trust.

At MicroStrategy, I designed a ReWOO plan-and-execute architecture with a ReAct fallback, then rebuilt AutoDash around a two-stage async pipeline. The aim was simple: make complex work easier to inspect, recover, and run with less waiting.

Memory + MCP

Context should survive the session without becoming a black box.

SimpleMem combines short- and long-term memory on ClickHouse and Redis, while MCP support gives agents a structured way to connect to tools such as Atlassian, GitHub, Notion, and Mosaic.

Retrieval

Useful retrieval needs more than a nearest-neighbour match.

For enterprise knowledge work, I paired vector RAG with a Neo4j knowledge graph, graph traversal, and entity matching. That hybrid approach cut query-resolution time by 70% for a Confluence knowledge hub.

Neural systems map.

An interactive topology of the engineering and applied-AI capabilities behind the systems I ship. Select a node to trace its production connections.

29 nodes  ·  31 synapses

A concise view of the work.

Software engineer with 3+ years building scalable backends and high-performance APIs in Python, Django, and FastAPI, now focused on applied GenAI: agentic workflows, RAG pipelines, MCP integrations, and evaluation frameworks. Currently AI Engineer at MicroStrategy, where I architect the agent stack behind AutoAgents and AutoDash.

Open résumé (PDF)
Current
AI Engineer 1

MicroStrategy Inc.

Impact
50%+

lower agent latency

Education
B.E., Information Technology

CGPA 9.57 / 10

Building something with AI? Let's talk.

Hiring, collaborating, or comparing notes on agent architecture. I read every message.

Contact form varun.motiyani14@gmail.com