PM-Agent-Plus

Inspiration

I saw first‑hand how much time teams waste on status meetings, risk checks, and manual plan updates. I wanted to build a system that treats AI as a proactive partner—automating routine PM tasks so real people can focus on strategy and execution.

What I Learned

• How to orchestrate multiple LLM‑based agents (planning, risk analysis, research) in a single workflow
  
• Best practices for combining LangChain/LangGraph with Google Gemini and OpenAI APIs
  
• Techniques for interactive network and Gantt visualizations using Plotly, NetworkX & PyVis
  
• Strategies for secure code execution in Docker‑backed AutoGen environments
  

How I Built It

I used FastAPI + Uvicorn for a REST backend, and Streamlit for the frontend. LangChain/LangGraph drives the Project Manager Assistant agent, while PyAutoGen coordinates a research‑team microservice in Docker. Task/dependency data lives in SQLite via SQLModel, and visualizations use Plotly and NetworkX/PyVis. Alerts and reports are generated with SendGrid, Google Calendar API, Pandoc and ReportLab.

Challenges Faced

• Ensuring agents never “hallucinate” dependencies or risks—solved via prompt‐anchoring and iterative validation
  
• Keeping long project plans in sync across multiple visualizations—addressed by a unified JSON model and event‑sourced logging
  
• Securely executing user‑defined code in research agents—implemented sandboxed Docker containers with strict resource limits
  

Built With

• autogen-(pyautogen)
• calendar
• docker
• fastapi
• google
• google?gemini-api
• langchain
• langgraph
• networkx
• openai-api
• pandoc
• plotly
• python
• pyvis
• reportlab
• sendgrid
• sqlite
• sqlmodel
• streamlit
• uvicorn