Skip to content

Repository files navigation

🤖 OnboardAI

Autonomous Client Onboarding — Powered by Agentic AI

Python Gemini License Hackathon

Built for the Agentic AI Hackathon 2026 · Team BinaryAgents


🧠 What is OnboardAI?

OnboardAI is a fully autonomous, multi-agent system that replaces manual client onboarding with intelligent, LLM-driven workflow execution.

Traditional onboarding is manual, slow, and error-prone.
OnboardAI replaces this with an LLM that thinks, decides, and acts — without human intervention.


⚡ Core Architecture

┌─────────────────────────────────────────────────────────┐
│                     OnboardAI System                      │
│                                                           │
│   Input (client data)                                     │
│       │                                                   │
│       ▼                                                   │
│  ┌─────────────┐   PLAN    ┌─────────────┐               │
│  │  🧠 Oogway  │ ────────► │  ⚡ Po      │               │
│  │  (Planner)  │           │  (Executor) │               │
│  │  Gemini LLM │ ◄──────── │  Tool Layer │               │
│  └─────────────┘  RESULT   └─────────────┘               │
│         │                        │                        │
│         │          DECISION      ▼                        │
│         └───────────────► ┌─────────────┐                │
│                            │  🛡️ Viper   │                │
│                            │  (Validator)│                │
│                            └─────────────┘                │
│                                  │                        │
│                    Continue / Retry / Abort               │
│                                  │                        │
│                                  ▼                        │
│                         outputs/report.md                 │
└─────────────────────────────────────────────────────────┘

🔁 The Agentic Loop

PLAN → ACTION → RESULT → DECISION → LOOP

Unlike a fixed pipeline, every step is decided dynamically by the LLM based on current state.
Remove the LLM and the system loses all intelligence — that's what makes it truly agentic.


👥 Agent Roles

Agent Character Role Technology
Planner 🧠 Oogway Decides the next action Gemini 2.5 Flash
Executor ⚡ Po Runs the tool for that action Python functions
Validator 🛡️ Viper Checks output, handles edge cases Rule-based logic

🛠️ Simulated Tools

Tool Simulates
send_email Welcome email dispatch
create_drive_folder Google Drive folder creation
create_notion_page Notion onboarding dashboard
create_airtable_record Airtable CRM record

⚠️ Edge Case Handling

Scenario System Response
Missing email Validator issues ABORT — onboarding halts safely
Duplicate client Validator issues WARN — logged, loop continues
Tool failure Validator issues RETRY — up to 2 retries per step
Invalid input Tool returns error, Validator escalates

📊 Structured Logging

Every step is fully transparent:

[PLAN]     LLM Reasoning: Email not yet sent and client has valid address.
[PLAN]     Next action selected: SEND_EMAIL
[ACTION]   Executor dispatching tool: SEND_EMAIL
[RESULT]   ✉️ Welcome email dispatched to TechGlobal <hello@techglobal.io>
[DECISION] Validator approved. → Step 'SEND_EMAIL' validated successfully.

📁 Project Structure

submission_BinaryAgents/
├── app/
│   ├── main.py                  # Entry point
│   ├── agents/
│   │   ├── planner_agent.py     # Oogway — Gemini LLM decisions
│   │   ├── executor_agent.py    # Po — tool dispatch
│   │   └── validator_agent.py   # Viper — result validation
│   ├── tools/
│   │   ├── email_tool.py
│   │   ├── drive_tool.py
│   │   ├── notion_tool.py
│   │   └── airtable_tool.py
│   ├── core/
│   │   ├── agent_loop.py        # PLAN → ACTION → RESULT → DECISION
│   │   └── state_manager.py     # Session state & history
│   └── utils/
│       ├── logger.py            # Color-coded structured logs
│       └── report_generator.py  # Markdown report writer
├── data/
│   └── sample_input.json        # Test client data (incl. edge cases)
├── docs/
│   ├── architecture.md
│   └── agent_flow.md
├── outputs/                     # Generated reports land here
├── requirements.txt
└── .env.example

🚀 Setup & Run

1. Clone and install

git clone <your-repo-url>
cd <repo-name>
pip install -r submission_BinaryAgents/requirements.txt

2. Configure API key

cp submission_BinaryAgents/.env.example submission_BinaryAgents/.env
# Edit .env and set your GOOGLE_API_KEY

3. Run

python submission_BinaryAgents/app/main.py

🧰 Tech Stack

Layer Technology
Language Python 3.10+
LLM Google Gemini 2.5 Flash (google-genai)
Config python-dotenv
Output Markdown (outputs/)
Architecture Custom multi-agent loop

🏆 Evaluation Criteria Alignment

Criterion How OnboardAI Addresses It
Autonomy (30%) LLM decides every step — no hardcoded sequence
Reasoning (25%) Planner returns explicit reasoning per decision
Reliability (25%) Retry, Warn, Abort tiers + heuristic fallback
Efficiency (20%) No sleep delays, minimal footprint, clean loop

👥 Team

Team BinaryAgents — Agentic AI Hackathon 2026 🚀

About

Autonomous agentic AI system that automates client onboarding using LLM-driven planning, tool execution, and decision loops.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages