Explore and develop deep agents with LangGraph
This repo is dedicated to hands-on experimentation with the LangGraph agent framework during October–November 2025. Projects are split into folders, each representing an independent agent implementation, demo, or prototype.
- All code is posted in its current state—raw, refactored, or exploratory.
- The focus is accelerated agent development, tool integration, and solutioning, rather than polished deliverables.
Background: Building on my experience as a production data scientist, ML engineer, and agent system experimenter (see main profile), I'm using this space for rapid prototyping and conceptual deep dives. My wider machine learning journey and skill development are documented in learning_journey, which shows my evolution from research scientist to full-stack ML system builder.
Basic_deepagent Project
This is a minimal, educational implementation of LangGraph's create_deep_agent function, designed to showcase the core ideas behind deep agents for complex, multi-step tasks. It demonstrates:
- Planning via automatic TODO lists
- Use of sub-agents for research and reflection (context specialization)
- A virtual file system for persistent memory and state
- Comprehensive system prompts for advanced reasoning
- The agent can search the web (using the Tavily API), plan, research, reflect, and iterate on results, demonstrating architecture and tool integration in a simple, learnable form. This project is best for learning or experimentation with deep agent patterns and LangGraph tooling.
Agent Project This folder contains multiple agent implementations, including the minimal_research_agent and academic_paper_tool subprojects:
- minimal_research_agent: A streamlined LangGraph ReAct agent that tests academic paper search and processing. It takes a research topic and question, finds and processes relevant academic papers, and uses a 9-stage pipeline to answer user queries with citations. It integrates with an academic paper tool for searching, downloading, processing, and querying papers, and provides detailed step-by-step agent reasoning.
- academic_paper_tool: (based on folder structure) Appears to be a modular set of scripts and configs for automated academic paper handling, likely supporting downloading, processing, and querying of academic literature, possibly for reuse by research-focused agents.
In summary, the Agent folder offers more specialized and modular agent implementations aimed at academic research and workflow automation.
Folders will expand as additional agent demos and experiments are added. Each represents a standalone investigation into agentic workflows, checkpointing, tool use, or persistent state logic according to the LangGraph paradigm.
- Growth mindset: Expect incomplete or unrefined code as a record of practical skill progression and iterative experimentation.
- Transparency: Projects are as-is, with evolutionary improvements over time.
- Documentation: Inline comments and markdown notes provided where possible; major learnings and design decisions will be chronicled in folder-specific
READMEs.
- ML engineers, agent system developers, and LangGraph enthusiasts seeking code examples, patterns, or inspiration.
- Anyone interested in the application and evolution of agentic AI architectures in Python.
- Main Profile: CJRockball — background, project links, portfolio
- Learning/Machine Learning Timeline: learning_journey — showcases Python/ML/AI skill progression, technical milestones, and project phases (see TIMELINE.md for details)
- Agent-Lab Companion Repo: agent-lab — reusable patterns and micro-services for agent systems
For inquiries, collaborations, or feedback:
You are welcome to use, contribute to, fork, or reference the materials herein in accordance with the repository license. 1