Minimal Workflow Engine – Tredence AI Engineering Assignment

This repository contains a minimal agent/workflow engine implemented in Python + FastAPI for the Tredence AI Engineering assignment.

The engine allows you to define:

• A set of nodes (execution steps)
• A shared state dictionary that flows through the nodes
• Edges between nodes (sequential or conditional)
• A simple looping mechanism
• A tool registry to plug in Python functions

It includes a working sample workflow (Option B):

Long text → split into chunks → generate chunk summaries → merge → refine summary in a loop.

---

Project Structure

tredence_assignment/
├── app/
│   └── main.py
├── README.md
└── requirements.txt

---

Getting Started

Install dependencies

pip install -r requirements.txt

Run the FastAPI app

uvicorn app.main:app --reload

Open Swagger UI:

http://127.0.0.1:8000/docs

---

Core Concepts

Node

A node represents a step in the workflow:

• id
• tool
• next
• branch condition
• conditional paths

State

A shared Python dictionary passed & updated through every node.

Tools

Registered using:

@register_tool("split_text")
def split_text_tool(state):
    ...

Conditions

Boolean functions for branching or looping.

Looping

Defined using LoopConfig:

• node_id
• condition
• max_iterations

---

Endpoints

POST /graph/create

Creates a workflow graph.

POST /graph/run

Runs a graph with initial state.

GET /graph/state/{run_id}

Returns the log and current state of a workflow run.

---

Sample Workflow (Option B – Summarization)

1. Splits text into chunks
2. Creates summaries
3. Merges summaries
4. Refines summary
5. Loops until summary meets limit

---

Future Improvements

• Database persistence
• WebSocket log streaming
• Parallel node execution
• Background task support
• Graph visualization UI

---

Summary

This project satisfies all required components:

• Minimal workflow engine
• State-driven execution
• Branching & looping
• Tool registry
• FastAPI backend
• Complete sample workflow