AutoAgentium

Welcome to AutoAgentium, my personal portfolio for building AI-powered multi‑agent workflows. Below you’ll find 5 bite‑sized subprojects, each one a mini project demonstrating a unique way to let these agents collaborate on real problems, from plotting stock charts to crafting full reports.

---

🛠️ Mini Projects

1. Multi-Agent Conversation(Comedy Duo)

Pattern: Here, I have taken a Two‑Comedian to chat.

• Actors: Cathy & Joe, each a ConversableAgent with custom system prompts.

• Flow: Joe kicks off with a punchline; Cathy builds on it. They riff back and forth until one says, “I gotta go.”

• Key features:

  • max_turns control
  • They exchange jokes, remembering punchlines and building off each other's humor.
  • Termination is dynamic: they stop when one says "I gotta go."
  • Tracks full conversation history, token usage, and allows you to generate summary reflections.

---

2. Sequential Chats and Customer Onboarding

Pattern: Step‑by‑Step Chat Pipeline

• Use case: Customer onboarding or ETL pipelines.

• How it works: Chain agents in a fixed WelcomeAgent → order—Greeter → DataCollector → Verifier, passing summaries or reflections as context.

• Benefits:

  • Modular design: Add or reorder steps by editing a list.
  • Flexible summaries: Quick carryover vs. thoughtful LLM reflections.

---

3. Reflection and Blogpost Writing

Pattern: Nested Chat Teams

• Manager agent spawns a mini‑team on demand.
• Example: A news micro‑team with research, drafting, and fact‑checking agents—all coordinated in one chat.
• Use case: Collaborative writing, turn‑based games or dynamic task delegation.

---

4. ChessMaster(Chess Game between 2 AI Agents)

Pattern: Tool-enabled agents with nested chats

• Agents: White & Black players using GPT-4 Turbo.
• Tool: BoardProxy executes get_legal_moves() and make_move() with python-chess.
• Flow: Player asks for moves, proxy returns legal UCI options, player chooses and proxy confirms.

• Why it's cool: Combines strategy, tool use, and turn-based coordination inside a conversational loop.

---

5. Coding and Financial Analysis

Pattern: Code generation + execution loop

• Roles: Code Writer drafts Python scripts; Executor runs them locally.
• Demo task: Fetch YTD(Year to Date) gains for NVDA/TSLA with yfinance and plot via matplotlib.

---

6. Stock Report Generation

Pattern: Group chat with planning and role distribution

• Team: Planner → Engineer → Executor → Writer → Admin.

• Workflow:

  1. Planner: Breaks down tasks: fetch, analyze, draft.
  2. Engineer: Writes the python code for each subtask.
  3. Executor: Runs that code and returns visuals, stats, etc.
  4. Writer: Drafts the final Markdown report
  5. Admin: checks progress, requests human approvals.

• Outcome: A complete stock report, with data pulled, charts generated, and blog written—all through agent interaction.

---

⚙️ Getting Started

1. Clone this repo:

   git clone https://github.com/maverick4code/AutoAgentium
   cd AutoAgentium

2. Install dependencies:

   pip install -r requirements.txt

---

If you have ideas to make this project better, want to collaborate, or just chat about AI automation, you can connect with me. LinkedIn: https://www.linkedin.com/in/sagar-shahari-703b84257/