OnboardAI — Autonomous Developer Onboarding Agent

NovaByte Technologies  ·  Syrus 2026  ·  Team Caffeine & Commit

---

📌 Problem Statement — PS-03

Organizations lack an autonomous onboarding system that can guide new software developers through company processes, personalize their onboarding experience based on role and experience, track completion, and generate structured confirmation to HR — all using a structured company knowledge base.

---

💡 What We Built

OnboardAI is an intelligent, agentic onboarding assistant that autonomously guides new developers at NovaByte Technologies through their entire onboarding journey — step by step, verified, and personalized.

It doesn't just answer questions. It plans, verifies, adapts, remembers, and acts — firing real integrations, growing an auto-generated knowledge base, scoring readiness, and notifying HR automatically on completion.

---

🎬 Demo & Presentation

Asset	Link
📹 Video Walkthrough	https://youtu.be/xhb4uEM2kGg
📊 Presentation Slides	https://drive.google.com/file/d/1jiV71kU49CwCPdyDc8lPhFUm-eTgfo8u/view?usp=drivesdk

---

✨ Core Features

🧠 Agent Intelligence

• Personalized onboarding paths based on role (Backend / Frontend / DevOps), level (Intern / Junior / Senior), tech stack, and team
• Step-by-step verified guidance — never marks a task complete without proof
  • Setup tasks → asks for terminal output (version numbers, logs)
  • Knowledge tasks → asks a quiz question from the actual KB docs
  • Attendance tasks → asks for ticket number or PR link
• Strike system — 3 failed verification attempts flags the task as TASK_STRUGGLE for HR review
• Adaptive guidance by level — Interns get copy-paste commands with explanations; Seniors get high-level peer-style guidance
• Proactive tips — context-aware tips delivered before the user asks, based on their role and team
• Context memory — remembers earlier struggles and references them later in the session

🔗 Integrations (MCP-Style 3-Tier Architecture)

Integration	Trigger	What Happens
💬 Slack	All 5 profile fields collected	Welcome message sent to #general, #onboarding, and team channel
🐙 GitHub	User requests access or reaches clone step	Repo access provisioned with clone instructions
📋 Jira	User requests Jira access	Board access granted with sprint and permission details
🎫 Starter Tickets	Checklist reaches ticket step	GitHub issue + Jira ticket auto-assigned with full acceptance criteria

All integrations are simulated (no real API keys required) — designed for demo purposes per hackathon constraints.

📚 Knowledge Base & RAG

• 11 structured resource files from PS-03 loaded and searchable
• Keyword-based RAG with source citations (from: filename.md)
• Auto-generated FAQs — every question asked during onboarding is saved to the KB with developer context, so future hires benefit automatically
• KB grows smarter with every onboarding session

💾 Memory & Persistence

• Session memory — progress saved to memory/ on every task completion
• Returning employees — automatically resume from where they left off with a personalised welcome-back message

📊 Scoring & HR Reporting

• Readiness Score — based on task completion, penalised for strikes and struggles
• Risk Score — flags if onboarding was rushed or suspiciously perfect
• Session summary auto-saved to session_logs/ with full breakdown
• HR email sent automatically on completion with name, role, team, completed tasks, pending tasks, scores, and flagged items

---

🏗️ Architecture

3-Tier Agent Design (MCP-Style)

Tier 1 — RAG (Always Searchable)
  company_overview · engineering_standards · architecture_documentation
  setup_guides · policies · org_structure · onboarding_faq

Tier 2 — Agent Logic (Read On-Demand)
  employee_personas → persona matching
  onboarding_checklists → checklist generation
  starter_tickets → ticket assignment

Tier 3 — Templates (Read Once at Generation)
  email_templates → HR completion email

Multi-Agent Router

User Message
    └─► Intent Router (Mistral)
            ├─► INFO_GATHERING   → Onboarding Concierge persona
            ├─► TECH_QUERY       → Senior Tech Lead persona
            ├─► TASK_COMPLETE    → QA Verifier persona
            └─► GENERAL          → General Mentor persona

File Structure

Syrus2026_caffeine_and_commit/
├── agent.py                  # Core agent — logic, memory, FAQ, scoring
├── app.py                    # Streamlit UI
├── rag.py                    # Keyword-based RAG system
├── email_sender.py           # HR email via Gmail SMTP
├── mock_slack.py             # Slack integration (simulated)
├── mock_jira.py              # Jira integration (simulated)
├── mock_github.py            # GitHub integration (simulated)
├── env_verifier.py           # Auto-verifies terminal output for setup tasks
├── onboarding_flow.py        # Structured Q&A intro flow
├── knowledge_base/           # 11 PS-03 resource files + auto-generated FAQs
├── session_logs/             # Per-session summaries with scores (.md)
├── memory/                   # Persistent employee progress (.json)
├── logo.png                  # App favicon and chat avatar
├── .env                      # API keys (not committed)
└── requirements.txt

---

🚀 Setup & Run

1. Clone the repository

git clone https://github.com/CMPN-CODECELL/Syrus2026_caffeine_and_commit
cd Syrus2026_caffeine_and_commit

2. Create and activate a virtual environment

python -m venv .venv

# Windows
.venv\Scripts\activate

# Mac / Linux
source .venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Configure environment variables

Create a .env file in the project root:

# Mistral AI
MISTRAL_API_KEY=your_mistral_api_key
MISTRAL_MODEL=mistral-small-latest

# Gmail SMTP (for HR email)
SENDER_EMAIL=your_gmail@gmail.com
SENDER_PASSWORD=your_gmail_app_password   # App Password, not account password
HR_EMAIL=hr@novabyte.com
SMTP_HOST=smtp.gmail.com
SMTP_PORT=587

Note on Gmail App Password: Go to Google Account → Security → 2-Step Verification → App Passwords. Generate one for "Mail".

5. Run the application

streamlit run app.py

---

📦 Dependencies

streamlit==1.43.0       # UI framework
mistralai==1.5.1        # LLM (Mistral AI)
chromadb==0.6.3         # Vector store
python-dotenv==1.0.1    # Environment variable management
Pillow                  # Logo/favicon support
# Email uses stdlib smtplib — no extra package needed

Install all at once:

pip install -r requirements.txt

---

🎬 Sample Demo Flow

1. User introduces themselves

   "Hi, I'm Riya Sharma. I'm a Backend Intern on the Payments team working with Node.js."

2. Agent detects persona → loads personalized 6-step checklist → fires Slack welcome automatically

3. Agent guides step-by-step with verification

   • Node.js setup → asks for node --version terminal output
   • Codebase walkthrough → asks a quiz question from architecture_documentation.md

4. Integrations fire automatically

   • GitHub repo access provisioned with clone instructions
   • Jira board access granted with sprint details

5. Starter ticket auto-assigned

   • GitHub issue #47 + Jira ticket PAY-247 assigned with acceptance criteria

6. On completion

   • Session summary saved to session_logs/
   • HR email sent with Readiness Score and Risk Score
   • Auto-generated FAQ added to knowledge_base/auto_generated_faqs.md

---

📊 Evaluation Criteria Coverage

Criteria	How We Address It
Agent Intelligence (30%)	Multi-agent router, verified task completion, adaptive guidance by level, proactive tips, strike system
Knowledge Retrieval (20%)	Keyword RAG with source citations, hallucination prevention via strict KB scope
System Architecture (20%)	3-tier MCP-style design, modular agent + integrations + RAG separation
Completion Reporting (10%)	Structured HTML/plain-text HR email with confidence + suspicion scores
Bonus Integrations (10%)	GitHub, Slack, Jira simulation; env auto-verifier; auto-generated FAQ KB

---

👥 Team — Caffeine & Commit

Name
Mahek Hingorani
Navya Rangwani
Khushi Dayaramani
Twinkle Gupta

---

🛠️ Tech Stack

Layer	Technology
LLM	Mistral AI (mistral-small-latest)
UI	Streamlit
RAG	Custom keyword-based retrieval
Email	Gmail SMTP (smtplib)
Integrations	Simulated Slack, GitHub, Jira
Storage	Local JSON (memory) · Markdown (session logs, FAQs)
Favicon & Avatar	Pillow (logo.png)

---

_{Built with ☕ and commits by Team Caffeine & Commit  ·  Syrus 2026  ·  NovaByte Technologies}