Poseidon-Lite

A demonstration of enterprise-grade orchestration for LLM and agentic systems.

Poseidon-Lite is an open demonstration of how to build and orchestrate a modular Retrieval-Augmented Generation (RAG) and Agentic LLM stack — without exposing proprietary data or internal logic.
It focuses on showing engineering design patterns, modular pipeline structure, and production readiness that mirror enterprise deployments.

---

🏗️ Architecture Overview

High-level flow:

Document ingestion → Text chunking → Embedding → Vector store → Retrieval → LLM Agent → Evaluation

yaml Copy code

Core modules:

• Pipelines — Handle ingestion, preprocessing, and embeddings.
• Agents — Orchestrate multi-step reasoning, retrieve context, and call tools.
• Evaluation — Validate retrieval accuracy and latency.
• Configs — Store environment and model parameters with clear separation of secrets.

---

🧩 Example Components

Module	Description	Key File
Embeddings	Converts text docs to vector embeddings using FAISS/Chroma	src/pipelines/embed_documents.py
Chunking	Splits text into semantically meaningful chunks	src/pipelines/chunker.py
Agent Orchestrator	Manages tool-calling and chain-of-thought	src/agents/orchestrator.py
Prompt Templates	Declarative prompt configurations	src/agents/prompts/
Evaluation	Benchmarking retrieval quality	src/evaluation/eval_metrics.py

---

🚀 Quickstart

git clone https://github.com/jherold2/Poseidon-Lite.git
cd Poseidon-Lite
pip install -r requirements.txt

# Set environment variables
cp configs/secrets_template.env .env

# Run embedding and retrieval demo
python src/pipelines/embed_documents.py
python notebooks/demo_agent.ipynb