In this project, we build an agentic workflow system.
pandas==2.2.3 openai==1.78.1 python-dotenv==1.1.0
├── workflow_agents/
│ ├── __init__.py ← (empty)
│ └── base_agents.py ← agent implementation file
├── direct_prompt_agent.py
├── augmented_prompt_agent.py
├── knowledge_augmented_prompt_agent.py
├── rag_knowledge_prompt_agent.py
├── evaluation_agent.py
├── routing_agent.py
└── action_planning_agent.py
* `workflow_agents` is a Python package containing all of agent class definitions.
* One script per agent to test their functionality has also been provided in the folder
## An OpenAI API is required to run the agent workflow.
## The Challenge: Building a Scalable Engine for Innovation
TPMs at InnovateNext Solutions are overburdened and face a significant bottleneck: due to their significant workload, turning the multiple product ideas they are handling, into actionable development plans is leading to miscommunications, varied output quality, and delays across all their projects. They need a foundational, AI-driven project management framework that can be applied company-wide.
## Project Steps:
First, we'll construct a robust library of diverse, reusable AI agents – the versatile building blocks for this and future advanced agentic systems. This is about crafting a core, adaptable toolkit.
Then, we'll deploy a selection of these agents to build the general-purpose agentic workflow for technical project management. We will demonstrate its power and flexibility by using their "Email Router" product specification (Product-Spec-Email-Router.txt) as the initial input for this pilot implementation.
The Audience for our solution are the technical project managers and the leadership team, particularly the Head of Product and Lead Technical Program Manager, at InnovateNext Solutions. They are looking for a robust system that not only works for the Email Router but also assists the technical project managers at InnovateNext for future product development.
Your Product: AI-Powered Agentic Workflow for Project Management (Pilot: Email Router)
Phase 1: The Agentic Toolkit
A Python package (workflow_agents) containing seven meticulously crafted and individually tested agent classes (base_agents.py):
DirectPromptAgent
AugmentedPromptAgent
KnowledgeAugmentedPromptAgent
RAGKnowledgePromptAgent (provided, but understand its role)
EvaluationAgent
RoutingAgent
ActionPlanningAgent
Standalone test scripts for each agent, proving their individual capabilities, along with screenshots of successful test runs.
Phase 2: The Project Management Workflow Implementation
A primary Python script (agentic_workflow.py) orchestrates a selection of the Phase 1 agents (ActionPlanningAgent, KnowledgeAugmentedPromptAgent, EvaluationAgent, RoutingAgent) to perform the multi-step technical project management task. This script will be designed as a general-purpose workflow.
For the pilot, this workflow will:
Accept a high-level prompt (simulating a TPM's request) and InnovateNext's Product-Spec-Email-Router.txt (as the example product spec).
Employ an Action Planning Agent to break down the overall goal into logical sub-tasks.
Utilize a Routing Agent to intelligently assign each sub-task to the appropriate specialized agent team.
Simulate a Product Manager team (a KnowledgeAugmentedPromptAgent for generating user stories based on the provided product spec, paired with an EvaluationAgent to ensure stories meet specific criteria like: "As a [type of user], I want [an action or feature] so that [benefit/value].").
Simulate a Program Manager team (a KnowledgeAugmentedPromptAgent for defining product features, paired with an EvaluationAgent to ensure features meet criteria like detailed descriptions, functionality, and user benefits).
Simulate a Development Engineer team (a KnowledgeAugmentedPromptAgent for creating detailed engineering tasks, paired with an EvaluationAgent to ensure tasks meet criteria like task IDs, descriptions, acceptance criteria, and estimations).
Produce a final, structured output representing the comprehensively planned project (for the Email Router), demonstrating the workflow's capability.
# Agentic-Workflows