GenAI Agents

The Problem

Teams that want to prototype generative‑AI agents often spend weeks stitching together LangChain, vector stores, and Flask glue code. The lack of ready‑made, end‑to‑end examples forces engineers to reinvent boiler‑plate for every new use case, slowing delivery and increasing bugs.

What This Does

GenAI_Agents is a curated collection of 52 Jupyter notebooks and supporting scripts that demonstrate concrete agent patterns—from a simple conversational bot (simple_conversational_agent.ipynb) to a multi‑agent workflow (multi_agent_collaboration_system.ipynb). Key assets include:

all_agents_tutorials/Weather_Disaster_Management_AI_AGENT.ipynb – the only notebook with an explicit main cell (line 447) that drives a small end‑to‑end run.
all_agents_tutorials/scripts/mcp_server.py – a Flask server that can expose any LangChain‑based agent as a REST endpoint.
requirements.txt – pins the core stack (flask, langchain, vector‑store libraries) used across all tutorials.

Each notebook bundles data files (e.g., data/vdb_entities.json) and visual assets, so a user can launch a demo with a single pip install -r requirements.txt and then run the notebook cells.

How It Is Wired

Execution starts in Weather_Disaster_Management_AI_AGENT.ipynb at the main cell. The cell calls two internal functions (search and agenerate) that appear 10 times each in the internal call graph, making them the highest‑traffic nodes. Both functions eventually invoke the generic _make_api_call helper (also called from 6 places) which performs the outbound LLM request.

The call chain looks like:

main (Weather_Disaster_Management_AI_AGENT.ipynb:447)
 └─> search  → _make_api_call → external LLM endpoint
 └─> agenerate → _make_api_call → external LLM endpoint

Other notebooks follow a similar pattern: they instantiate a LangChain AgentExecutor (called from 5 places) and then invoke helper utilities such as save_file, show_md_file, and chunk_large_text (each called from 6 places) to persist intermediate results.

File‑level responsibilities (ranked by call volume) include:

File	Core duties	External effect
`Academic_Task_Learning_Agent_LangGraph.ipynb`	Configures API keys, reduces dicts, runs `agenerate`	Calls LLM
`database_discovery_fleet.ipynb`	Opens SQLite (`chinook.db`), runs queries, builds prompts	DB read/write + LLM
`contextual_quoting_agentic_system.ipynb`	Updates workflow state, handles tool errors, writes to DB	DB write
`Gutenbergs_Sage.ipynb`	Builds vector store, runs external `ollama` command	External command + LLM + network
`EU_Green_Compliance_FAQ_Bot.ipynb`	Retrieves relevant chunks, scores relevance	LLM + network

No circular import cycles were detected; the three Python modules (mcp_server.py, e2e_testing_agent_app.py, extract_items.js) are isolated, each with 0 import edges, which simplifies refactoring.

How To Use It

# Clone the repo (exact URL required)
git clone https://github.com/moses-y/GenAI_Agents
cd GenAI_Agents

# Install the Python stack
pip install -r requirements.txt

No additional configuration files are present; notebooks expect the user to set API keys in‑notebook (e.g., via configure_api_keys). To run a demo, open the desired notebook in JupyterLab and execute the cells, or start the Flask server for a notebook that exports an endpoint:

python all_agents_tutorials/scripts/mcp_server.py

The server will listen on the default Flask port and forward requests to the underlying LangChain agent defined in the notebook.

Real‑World Use

A product team could copy project_manager_assistant_agent.ipynb, replace the placeholder prompts with their own project data, and embed the resulting AgentExecutor in a Flask service (mcp_server.py). The service would then accept JSON payloads describing new tasks and return a structured task‑dependency graph, leveraging the same LLM call path (_make_api_call) used across the repo.

Code Health & Issues

MEDIUM – Enable Dependabot or Renovate – only one manifest (requirements.txt); no bot configured.
MEDIUM – Strip notebook outputs – all_agents_tutorials/Weather_Disaster_Management_AI_AGENT.ipynb contains >5 MB of stored output.
MEDIUM – Split oversized notebook cell – all_agents_tutorials/business_meme_generator.ipynb has a cell >8 k characters.
MEDIUM – Move large binaries to Git LFS – images/substack_image.png (10.5 MB) and two JSON blobs >7 MB each.
MEDIUM – Set a random seed in stochastic notebooks – business_meme_generator.ipynb runs sampling without a fixed seed.

Additional observations from the repository hygiene report: tests exist (7 files) but no CI pipeline, no lockfile, and a license file is present.

The Bottom Line

GenAI_Agents offers a breadth of ready‑to‑run agent examples that can accelerate prototyping, with clear entry points and reusable helper functions. The code base lacks production‑grade scaffolding (CI, lockfile, LFS for large assets), so teams should treat it as a learning sandbox and add the missing DevOps pieces before shipping. Suitable for engineers who need fast, concrete references for LangChain‑based agents and are comfortable cleaning up notebook artifacts.