Mohammad Daoud Farooqi
Partner Solutions Architect, AI at MongoDB
AI agents in production are stitched together from four systems that were never meant to talk to each other: a vector database, a keyword engine, a graph store, and a system of record. My job is to collapse that into one. I help banks, insurers, and large data teams ship agents in the industries where “move fast and break things” was never an option.
Ai4 2026 · The Venetian, Las Vegas
For Agents: Deployment at Scale
Collapsing the Agent Stack with MongoDB + Mastra
Production AI agents need three things at runtime: vectors for retrieval, durable state for workflows, and long-term memory. Most teams assemble these from separate systems (a vector database, a cache, a relational store, and an orchestration framework) and watch cost and operational overhead scale linearly as they grow. In this joint session, MongoDB and Mastra show how to collapse that sprawl onto a single platform: building a TypeScript agent on Mastra backed by MongoDB Atlas, with vectors, state, and memory in one cluster under one security and operational model. We'll co-build live and walk through the real trade-offs between performance, latency, and operational expense at scale.
What you'll take away
- A working reference architecture: a TypeScript agent built on Mastra, backed by a single MongoDB Atlas cluster for vectors, durable state, and long-term memory.
- The real trade-offs between performance, latency, and operational expense as agent workloads scale.
- The open-source patterns behind the demo: memory services, hybrid retrieval, and agentic quickstarts you can run today.
About
Collapsing the Agent Stack
I'm a Partner Solutions Architect for AI at MongoDB and the author of the technical whitepaper on Constitutional AI ethical governance with MongoDB Atlas. My work is fixing a problem most teams know too well: AI agents in production are stitched together from four systems that were never meant to talk to each other. I help enterprises move agents into production in the industries where reliability, governance, and compliance are non-negotiable.
I came from financial services, so I lived on the customer side of data governance, security, and compliance long before I had to architect around them. Today I build MongoDB's Database-as-a-Service platform and the open-source AI reference architectures that run on top of it: long-term memory, hybrid vector and full-text search, and agentic quickstarts adopted across the partner ecosystem.
Beyond the Constitutional AI whitepaper, I wrote MongoDB's guide to building AI memory systems with AWS and Claude, and most recently an official Anthropic cookbook, published in anthropics/claude-cookbooks, showing MongoDB Atlas as the retrieval, graph, and system-of-record layer for Claude Managed Agents. At Ai4, I'll show how to collapse the modern agent stack, vector search, memory, and conversational state, into a single operational data layer.
Focus Areas
Writing
Publications
Published technical writing on governing, remembering, and retrieving for production AI systems
Constitutional AI: Ethical Governance with MongoDB Atlas
A complete guide to building self-governing AI systems on Atlas: RLAIF self-critique, hybrid search with two-stage retrieval and reranking, Change Streams audit trails, and Queryable Encryption. Written for regulated finance, healthcare, and legal.
Read →Build AI Memory Systems with MongoDB Atlas, AWS, and Claude
Long-term memory for AI agents on Atlas, AWS Bedrock, and Claude: hierarchical memory structures, importance scoring, and semantic recall. Pairs with the open-source ai-memory service below.
Read →The Converged Datastore for Agentic AI
Why agent runtimes collapse onto one operational datastore instead of a sprawl of vector databases, caches, and relational stores. The thesis behind the Ai4 session.
Read →Fraud Review Agent with MongoDB Atlas and Claude Managed Agents
Published in Anthropic’s official cookbook repository: MongoDB Atlas as the retrieval engine, graph store, and system of record for a human-in-the-loop fraud-review agent. Covers all four retrieval patterns ($vectorSearch, $search, $rankFusion, $graphLookup), a credential-safe host-side data path, and audit persistence.
Read →I also write about AI systems, MongoDB, and system design on my personal blog, threadwaiting.com →
Open Source
Agent Infrastructure You Can Run Today
Reference architectures and developer tooling for production AI agents, built at MongoDB and adopted across the partner ecosystem
ai-memory
AI Memory Service: long-term memory for agents on MongoDB Atlas and AWS Bedrock. Hierarchical memory, importance scoring, semantic search, and conversation summarization.
mongodb-llama-stack
MongoDB integration for Llama Stack: vector, full-text, hybrid search, and graph-enhanced retrieval built on native Atlas features. Registered upstream as an external provider in OGX, formerly Meta’s Llama Stack.
memory-mcp
MemoryMCP: persistent memory, semantic caching, and hybrid MongoDB search for AI assistants via the Model Context Protocol.
mongodb-mastra-agentic-ai-qs
Fraud Investigation Console built on MongoDB and Mastra: a TypeScript agent with vectors, state, and memory in one Atlas cluster. The pattern behind the Ai4 session.
maap-anthropic-qs
End-to-end MAAP stack combining MongoDB Atlas with Anthropic’s Claude models for an agentic conversational interface. A production-shaped quickstart.
mongodb-gemini-extension
The official Gemini CLI extension for MongoDB, wrapping the mongodb-mcp-server npm package for one-command installation with automatic upstream updates.
More Reference Architectures
Platform Work at MongoDB
Internal platform engineering, not open source, but very much production.
MongoDB Database-as-a-Service Platform
The platform layer that provisions and runs MongoDB as a managed service, plus the control plane automating provisioning, lifecycle, and tenant orchestration for partners and OEMs.
Partner Pavilion
A full-stack production storefront and SSO admin dashboard showcasing partner AI, streaming, workflow-orchestration, and semantic-search integrations with Atlas.
Industrial Predictive Maintenance
Predictive maintenance system for industrial operations built on AWS, MongoDB Atlas, and Voyage AI embeddings.
Speaking & Judging
On Stage
Talks and judging on production agent infrastructure, AI governance, and the converged data layer
Ai4 2026 — For Agents: Deployment at Scale
Collapsing the agent stack with MongoDB + Mastra. Joint session with Mastra, co-built live on stage. Session details
Google Cloud Rapid Agent Hackathon
Judge, MongoDB Track
Judged Google’s global hackathon on building agents for real-world challenges: 14,000+ participants and $60,000 in prizes, with agents built on Gemini, Google Cloud Agent Builder, and partner MCP servers.
View on Devpost →Agentic Apps on Real-Time Data with MongoDB Atlas & Confluent
Public co-presented webinar on closing the loop between live streams and agents: Kafka and Flink for real-time anomaly detection, Atlas Vector Search for context, and Atlas Stream Processing feeding actions back into the system.
MongoDB Developer Day at a Global Investment Bank
Hands-on developer day with a runnable cookbook: Atlas AI workloads, LangChain and LangSmith agent patterns, AgentCore, Claude Code, and the MongoDB MCP server.
Build for the Agentic Era
Public joint webinar with LangChain on the future of LLM application development for a leading US grocery retailer: MongoDB for agentic apps, the agent development lifecycle, and a live MongoDB + LangSmith demo.
From Prompt to Production: Using MongoDB Agent Skills with Vibe Coding Tools
Public webinar on going from an idea in a prompt to a production-ready, governed AI workflow with MongoDB Agent Skills, MCP, and vibe coding tools, delivered to a 400+ audience.
View event →The Converged Datastore: Agentic for Insurance & Other Industries
Featured speaker in front of 500+ developers.
Booking talks on production agents, AI memory, and governance for regulated industries. Get in touch
Contact
Let's Talk Agents
Attending Ai4? Exploring a partnership or a speaking slot? I'd love to hear from you.
farooqimdd@gmail.com
linkedin.com/in/farooqimdd
GitHub
github.com/mohammaddaoudfarooqi
Phone
+91-9599555734
Launch a chat →
At Ai4 2026?
I'll be at The Venetian, Las Vegas, August 5–7. If you're building agents for a regulated industry, or just want to talk shop about memory, retrieval, and state, pick “Meet at Ai4 2026” in the form and let's find a time.