Add MongoDB Text-to-MQL Agent Tutorial to Gen-AI Showcase

[MMBazel]

Add MongoDB Text-to-MQL Agent Tutorial to Gen-AI Showcase

This PR adds a comprehensive tutorial showcasing how to build natural language agents using MongoDB Atlas and LangChain's Text-to-MQL tools. The guide walks developers through two agent architectures:

• ReAct-style agent – ideal for fast prototyping, demos, and dynamic tool usage.
• LangGraph-style agent – designed for production with deterministic, auditable flows.

Key Features

• Use the langchain_mongodb.agent_toolkit to generate, validate, and execute secure MongoDB aggregation pipelines from plain English.
• Simplified prompt design using MONGODB_AGENT_SYSTEM_PROMPT.
• Detailed comparison of ReAct vs. graph-style agents (speed, accuracy, tradeoffs).
• Patterns for extending agents with memory, hybrid search (Vector + Full-Text + MQL), and observability.

Audience

This tutorial is intended for developers and teams building:

• GenAI agents over structured data
• Internal tools and analytics bots
• Production-grade agentic systems with audit/control requirements

[ajosh0504]

Nice work! I like the comparison between an easy (React) and a more customized (LangGraph) agent. Few comments:

• I found the wording "graph-based agent" a bit confusing. I took it to mean it was using graph-based retrieval or GraphRAG. Might want to specifically say "custom LangGraph agent" or similar.
• The long outputs at every step make the notebook a bit hard to read. Consider clearing the outputs of some cells
• In the LangGraph agent, curious why there is a conditional edge after generate_query? Shouldn't check_query, run_query, etc., always be run if the agent decides to generate a query?
• Instead of saving the output to a local file, you could persist it in MongoDB? Could also use the MongoDB LangGraph checkpointer to show how to use these persisted outputs as memory for the agent.

[MMBazel]

Nice work! I like the comparison between an easy (ReAct) and a more customized (LangGraph) agent. Few comments:

• I found the wording "graph-based agent" a bit confusing. I took it to mean it was using graph-based retrieval or GraphRAG. Might want to specifically say "custom LangGraph agent" or similar.
• The long outputs at every step make the notebook a bit hard to read. Consider clearing the outputs of some cells.
• In the LangGraph agent, curious why there is a conditional edge after generate_query? Shouldn't check_query, run_query, etc., always be run if the agent decides to generate a query?
• Instead of saving the output to a local file, you could persist it in MongoDB? Could also use the MongoDB LangGraph checkpointer to show how to use these persisted outputs as memory for the agent.

V2 Changes & Updates Made

1) Functionality added

• Persistent Memory: Replaced file-based saving with MongoDBSaver (LangGraph’s built-in checkpointer) and added an LLMSummarizingMongoDBSaver class that generates human-readable step summaries via the LLM.
• Quick Reference & Smoke Test: Appended a System Initialization & Quick Reference section and a if __name__ == "__main__": test_enhanced_summarization() smoke-test snippet for one-click validation.
• Enhanced Memory Management: Improved inspection and cleanup utilities—inspect_thread_with_summaries_enhanced, list_conversation_threads, clear_thread_history, and memory_system_stats—for better observability and maintenance.
• Interactive CLI: Refined interactive_query() with clear commands (exit, threads, switch, debug) and seamless thread switching to support live exploratory sessions.

2) Workflow restructuring

• Introduction overhaul: Reorganized front-matter into logical sections—Overview, Use Cases, Business Applications, Technical Components, Prerequisites, and Network Setup—to guide readers step by step.
• Error-handling cleanup: Converted every bare except: into except Exception as e: and surfaced exception details in logs.
• Control-flow clarification:
  • Rationale for keeping the conditional edge:
    1. Future-proofing: If the LLM ever returns a non-tool response (e.g. a clarification request or “I don’t know”), the branch cleanly skips validation and execution without graph rewrites.
    2. Semantic clarity: Explicitly encodes “only validate/execute when a query was generated,” making the intent and node roles crystal clear.
    3. Scalable branching: Allows easy addition of fallback or help branches later by adding new conditional paths from generate_query rather than rewiring hard-coded edges.

3) New uses

• Business applications: Added real-world query examples (analytics, recommendations, trend analysis, geographic analysis) to showcase practical value.

• Demos & agent comparison: Structured demo functions (demo_basic_queries, demo_conversation_memory, compare_agents_with_memory) under clear headings, with usage instructions and side-by-side performance analysis.

• Comprehensive test suite: Introduced run_comparison_tests() to automatically run simple, moderate, and complex query scenarios across both agents.

• Interactive exploration: Enhanced live notebook usage via the improved interactive_query() interface, encouraging hands-on data exploration and debugging.

[Copilot]

Copilot reviewed 1 out of 1 changed files in this pull request and generated no comments.