GitHub - vaishnaviwangalwar-cpu/Chatbot-Agent-and-Multi-Agent-RAG-: A Two-project workshop covering AI chatbots, multi-agent systems, RAG pipelines, prompt engineering, and cloud deployment — all built with production-grade architecture.

From Zero to Production-Ready AI Applications

A hands-on, two-project workshop covering AI chatbots, multi-agent systems, RAG pipelines, prompt engineering, and cloud deployment — all built with production-grade architecture.

🚀 Quick Start · 📦 Project 1: ChatBot · 🧠 Project 2: Multi-Agent RAG · ☁️ Deployment · 📚 Workshop Modules

🗺️ Workflow

mindmap
  root((🎓 AI Workshop))
    🤖 Project 1
      ChatBot Agent
      Prompt Engineering
      Tool Calling
      Session Memory
      Streaming SSE
    🧠 Project 2
      Multi-Agent RAG
      Orchestrator Agent
      Parsing Agent
      Vector Embeddings
      ChromaDB
    ☁️ Deployment
      Docker
      Cloud Run
      Secret Manager
    💡 Bonus
      CI/CD
      Monitoring
      Cost Optimization

✨ Highlights

🏗️ Production Architecture SOLID & DRY principles throughout Pydantic schemas for type safety Modular service layers Environment-based configuration	🧪 Hands-On Learning Copy-paste ready code snippets Step-by-step build instructions Interactive Web UI included Swagger API docs auto-generated
🔧 Real-World Tools Google Gemini 3.1 SDK ChromaDB vector database Server-Sent Events streaming Docker containerization	🚀 Deploy Anywhere Local development server Docker container builds Google Cloud Run Secret Manager integration

📦 Repository Structure

clg-ai-workshop/
│
├── 📁 chatbot_agent/              ← Project 1: AI ChatBot Agent
│   ├── app/
│   │   ├── main.py                # FastAPI app setup & health check
│   │   ├── config.py              # Pydantic BaseSettings (.env reader)
│   │   ├── schemas/chat.py        # Request/Response models
│   │   ├── prompts/
│   │   │   ├── system_prompts.py  # 4 prompt engineering techniques
│   │   │   └── campus_faqs.py     # Knowledge base dataset
│   │   ├── tools/campus_tools.py  # DateTime, Calculator, FAQ Lookup
│   │   ├── services/
│   │   │   ├── memory_service.py  # Sliding-window session manager
│   │   │   └── agent_service.py   # Gemini SDK orchestration
│   │   └── api/chat_router.py     # REST API endpoints
│   ├── static/                    # Glassmorphic Web UI
│   ├── Dockerfile
│   ├── requirements.txt
│   └── README.md
│
├── 📁 multi_agent_rag/            ← Project 2: Multi-Agent RAG
│   ├── app/
│   │   ├── main.py                # FastAPI app with lifespan events
│   │   ├── config.py              # Centralized configuration
│   │   ├── genai_client.py        # Shared GenAI client (DRY)
│   │   ├── prompts/
│   │   │   ├── routing_prompt.py  # Intent classification
│   │   │   ├── chat_prompt.py     # Casual conversation handler
│   │   │   └── rag_prompt.py      # XML grounded answer generation
│   │   ├── schemas/rag.py         # Pydantic schemas
│   │   ├── services/
│   │   │   ├── orchestrator_agent.py    # Central coordinator
│   │   │   ├── parsing_agent.py         # PDF/Image text extraction
│   │   │   ├── rag_service.py           # Retrieve → Augment → Generate
│   │   │   └── vector_store_service.py  # ChromaDB operations
│   │   └── api/rag_router.py      # REST endpoints
│   ├── sample_docs/               # Seed policy documents
│   ├── static/                    # Web UI
│   ├── Dockerfile
│   ├── requirements.txt
│   └── README.md
│
└── Closing.md                     # Deployment & bonus topics

🚀 Quick Start

Prerequisites

Tool	Version	Purpose
🐍 Python	3.10+	Runtime
🔑 Google AI Studio	—	Get your Gemini API Key
🐳 Docker	Latest	(Optional) Containerization
☁️ Google Cloud SDK	Latest	(Optional) Cloud Run deployment

⚡ Project 1 — ChatBot Agent Setup

1️⃣ Clone & Navigate

git clone https://github.com/vaishnaviwangalwar-cpu/clg-ai-workshop.git
cd clg-ai-workshop/chatbot_agent

2️⃣ Create Virtual Environment

python -m venv venv

# macOS/Linux
source venv/bin/activate

# Windows
venv\Scripts\activate

3️⃣ Install Dependencies

pip install -r requirements.txt

4️⃣ Configure Environment

cp .env.example .env

Edit .env with your credentials:

GEMINI_API_KEY=your_google_ai_studio_api_key_here
GEMINI_MODEL=gemini-3.1-flash-lite
APP_ENV=development
HOST=0.0.0.0
PORT=8000
MAX_SESSION_TURNS=10

5️⃣ Launch!

uvicorn app.main:app --reload --port 8000

Endpoint	URL
🌐 Web UI	http://localhost:8000
📖 Swagger Docs	http://localhost:8000/docs
💚 Health Check	http://localhost:8000/health

⚡ Project 2 — Multi-Agent RAG Setup

1️⃣ Navigate to Project

cd clg-ai-workshop/multi_agent_rag

2️⃣ Create Virtual Environment

python -m venv venv

# macOS/Linux
source venv/bin/activate

# Windows
venv\Scripts\activate

3️⃣ Install Dependencies

pip install -r requirements.txt

4️⃣ Configure Environment

cp .env.example .env

Edit .env with your credentials:

GEMINI_API_KEY=your_google_ai_studio_api_key_here
GEMINI_MODEL=gemini-3.1-flash-lite
EMBEDDING_MODEL=gemini-embedding-001
CHROMA_PATH=./chroma_db
TOP_K=3

5️⃣ Launch!

uvicorn app.main:app --reload --port 8001

Endpoint	URL
🌐 Web UI	http://localhost:8001
📖 Swagger Docs	http://localhost:8001/docs
💚 Health Check	http://localhost:8001/health

📦 Project 1 — ChatBot Agent

A production-grade AI ChatBot with prompt engineering, tool calling, session memory, and streaming responses.

🎯 What You'll Learn

graph LR
    A[📝 Prompt<br>Engineering] --> B[🔧 Tool<br>Calling]
    B --> C[🧠 Session<br>Memory]
    C --> D[⚡ Streaming<br>SSE]
    D --> E[🎨 Web<br>UI]

    style A fill:#6C5CE7,stroke:#5A4BD1,color:#fff
    style B fill:#00B894,stroke:#00A381,color:#fff
    style C fill:#FDCB6E,stroke:#E6B85C,color:#333
    style D fill:#E17055,stroke:#CA6349,color:#fff
    style E fill:#0984E3,stroke:#0876CC,color:#fff

📝 Module 3 — Prompt Engineering Techniques

The chatbot implements 4 distinct prompt engineering strategies that you can switch between at runtime:

#	Technique	Description	Best For
1	Standard Baseline	Minimal system prompt	Simple Q&A
2	Structured XML Tags	XML-formatted instruction blocks	Precision & control
3	Few-Shot Examples	Input/output demonstration pairs	Consistent formatting
4	Chain-of-Thought	Step-by-step reasoning guidance	Complex problem solving

💡 Try it: Switch between prompt strategies in the Web UI dropdown to see how each one changes the model's behavior and response quality!

🔧 Tool Calling (Function Calling)

The chatbot agent is equipped with 3 built-in tools that Gemini can invoke automatically:

Tool	Function	Example Query
🕐 `get_current_datetime`	Returns current date & time	"What time is it?"
🧮 `calculate`	Safe mathematical expression evaluator (AST-based)	"What's 15% of 2400?"
📋 `lookup_faq`	Searches campus knowledge base	"What's the library timing?"

User: "What's 25 * 48 + 300?"
├── 🤖 Gemini detects math intent
├── 🔧 Calls calculate("25 * 48 + 300")
├── 📊 Tool returns: 1500
└── 💬 Agent: "The result of 25 × 48 + 300 is **1,500**"

🧠 Session Memory Management

Conversation context is managed via an in-memory sliding window:

┌─────────────────────────────────────────┐
│         Session Memory (MAX_TURNS=10)    │
├─────────────────────────────────────────┤
│  Turn 1: User → "Hi!"                  │  ← Oldest
│  Turn 2: Bot  → "Hello! How can I..."  │
│  Turn 3: User → "Tell me about..."     │
│  ...                                    │
│  Turn 10: Bot → "Here's what I found"  │  ← Newest
├─────────────────────────────────────────┤
│  Turn 11 arrives → Turn 1 evicted 🗑️   │
└─────────────────────────────────────────┘

No database required — pure in-memory sliding window
Caps token usage — prevents context window overflow
Session isolation — each session ID maintains its own history

⚡ Dual Response Modes

Mode	Endpoint	Transport	Use Case
Standard	`POST /api/chat`	JSON response	Simple integrations
Streaming	`POST /api/chat/stream`	Server-Sent Events (SSE)	Real-time UI updates

Both modes share the same session memory — switching modes mid-conversation preserves full context.

🧠 Project 2 — Multi-Agent RAG Assistant

A multi-agent Retrieval-Augmented Generation system with document ingestion, vector search, and grounded answers.

🏗️ Architecture & Data Flow

graph TD
    User([👤 User / Web UI]) -->|1. Ask Query or Upload File| Orchestrator[🎯 Orchestrator Agent]

    subgraph " 🔍 Query Pipeline"
        Orchestrator -->|2a. Classify intent via Gemini| Orchestrator
        Orchestrator -->|2b. Route to RAG Agent| RAG[📚 RAG Agent]
        RAG -->|3. Embed query & search| DB[(🗄️ ChromaDB)]
        DB -->|4. Return Top-K chunks| RAG
        RAG -->|5. Build XML grounded prompt| GenLLM(🤖 Gemini API)
        GenLLM -->|6. Return grounded answer| RAG
        RAG -->|7. Return answer + sources| Orchestrator
    end

    subgraph " 📄 Ingestion Pipeline"
        Orchestrator -->|2c. Delegate file parsing| Parser[📄 Parsing Agent]
        Parser -->|3. Gemini multimodal OCR| ParserLLM(🤖 Gemini API)
        ParserLLM -->|4. Return markdown text| Parser
        Parser -->|5. Return parsed text| Orchestrator
        Orchestrator -->|6. Save & embed| VectorService[📊 Vector Store]
        VectorService -->|7. Request embeddings| EmbeddingsAPI(🔢 Embeddings API)
        EmbeddingsAPI -->|8. Return vectors| VectorService
        VectorService -->|9. Store chunks & vectors| DB
    end

    Orchestrator -->|📨 Final Response| User

    style User fill:#6C5CE7,stroke:#5A4BD1,color:#fff
    style Orchestrator fill:#00B894,stroke:#00A381,color:#fff
    style RAG fill:#0984E3,stroke:#0876CC,color:#fff
    style Parser fill:#FDCB6E,stroke:#E6B85C,color:#333
    style DB fill:#E17055,stroke:#CA6349,color:#fff
    style GenLLM fill:#A29BFE,stroke:#9088E4,color:#fff
    style ParserLLM fill:#A29BFE,stroke:#9088E4,color:#fff
    style VectorService fill:#FD79A8,stroke:#E46D96,color:#fff
    style EmbeddingsAPI fill:#74B9FF,stroke:#67A6E5,color:#333

🎯 What You'll Learn

🎯 Multi-Agent Coordination

The system uses a central orchestrator pattern where every request flows through a coordinator agent:

User Query → Orchestrator Agent
                ├── Intent: "policy_question" → RAG Agent → Grounded Answer
                ├── Intent: "casual_chat" → Direct LLM Response
                └── Intent: "file_upload" → Parsing Agent → Vector Store

The Orchestrator uses Gemini to classify intent before routing, ensuring each specialist agent only handles what it's designed for.

📄 Multimodal Document Ingestion

The Parsing Agent supports 3 document formats — no local parsing libraries needed:

Format	Method	How It Works
📄 `.txt`	Direct read	Plain text extraction
📕 `.pdf`	Gemini Vision	Multimodal LLM reads pages as images
🖼️ Images	Gemini Vision	OCR via `gemini-3.1-flash-lite`

🔥 Key Insight: Instead of using traditional PDF parsers like PyPDF2 or pdfplumber, this project leverages Gemini's multimodal capabilities to extract layout-aware text directly from document images!

🔢 Vector Embeddings & Search

Document → Chunk (paragraphs) → Embed (gemini-embedding-001) → Store (ChromaDB)
                                                                      ↓
User Query → Embed → Cosine Similarity Search → Top-K Chunks → Grounded Answer

Component	Technology	Purpose
Embeddings	`gemini-embedding-001`	Convert text to high-dimensional vectors
Vector DB	ChromaDB (persistent)	Store & query vectors by similarity
Retrieval	Cosine similarity	Find most relevant document chunks
Top-K	Configurable (default: 3)	Number of chunks to retrieve

🛡️ Grounded Answer Generation

Retrieved chunks are injected into a strict XML prompt template that enforces grounding rules:

<context>
  <document source="policy.txt" chunk="3">
    Students must maintain a minimum 75% attendance...
  </document>
</context>

<rules>
  - ONLY answer based on the provided context
  - If the context doesn't contain the answer, say so clearly
  - Cite the source document for every claim
</rules>

🛡️ This structured grounding approach prevents hallucination by constraining the model to only use information present in the retrieved documents.

📚 Workshop Modules

Module	Topic	Project	Key Concepts
1	🏗️ Architecture & Setup	Both	SOLID principles, project scaffolding, virtual environments
2	🔌 Google GenAI SDK	Both	API keys, model configuration, SDK initialization
3	📝 Prompt Engineering	ChatBot	XML tags, few-shot, chain-of-thought, baseline comparison
4	🧠 Memory & Sessions	ChatBot	Sliding window context, session isolation, token management
5	🔧 Tool / Function Calling	ChatBot	Python tool definitions, automatic invocation, result formatting
6	⚡ Streaming (SSE)	ChatBot	Server-Sent Events, real-time token delivery, dual modes
7	📄 Document Parsing	Multi-Agent RAG	Multimodal PDF/image extraction via Gemini Vision
8	🔢 Embeddings & Vectors	Multi-Agent RAG	`gemini-embedding-001`, chunking strategies, ChromaDB
9	🎯 Multi-Agent Orchestration	Multi-Agent RAG	Intent classification, agent routing, specialist coordination
10	🛡️ RAG Pipeline	Multi-Agent RAG	Retrieve → Augment → Generate, grounding, source citation
11	☁️ Deployment	Both	Docker, Cloud Run, Secret Manager, public URLs

☁️ Deployment

🐳 Docker Build & Run

ChatBot Agent:

cd chatbot_agent
docker build -t chatbot-agent .
docker run -p 8000:8000 --env-file .env chatbot-agent

Multi-Agent RAG:

cd multi_agent_rag
docker build -t multi-agent-rag .
docker run -p 8001:8001 --env-file .env multi-agent-rag

☁️ Google Cloud Run Deployment

1. Authenticate & Set Project

gcloud auth login
gcloud config set project YOUR_PROJECT_ID

2. Store API Key in Secret Manager

echo -n "YOUR_GEMINI_API_KEY" | \
  gcloud secrets create gemini-api-key --data-file=-

3. Push to Artifact Registry

gcloud builds submit --tag gcr.io/YOUR_PROJECT_ID/chatbot-agent

4. Deploy to Cloud Run

gcloud run deploy chatbot-agent \
  --image gcr.io/YOUR_PROJECT_ID/chatbot-agent \
  --platform managed \
  --region us-central1 \
  --allow-unauthenticated \
  --set-secrets="GEMINI_API_KEY=gemini-api-key:latest"

📖 See detailed deployment guides: chatbot_agent/DEPLOYMENT.md | multi_agent_rag/DEPLOYMENT.md

🛠️ Tech Stack

Layer	Technology	Role
🌐 Backend	FastAPI	Async REST API framework
🤖 AI/LLM	Google Gemini 3.1 (`google-genai`)	Language model & multimodal AI
🔢 Embeddings	`gemini-embedding-001`	Text-to-vector conversion
🗄️ Vector DB	ChromaDB	Persistent vector storage & search
✅ Validation	Pydantic v2	Schema validation & settings management
🎨 Frontend	HTML5 / CSS3 / JavaScript	Glassmorphic interactive Web UI
🐳 Container	Docker	Production containerization
☁️ Cloud	Google Cloud Run	Serverless container hosting
🔐 Secrets	GCP Secret Manager	Secure credential management

🔌 API Reference

ChatBot Agent Endpoints

Method	Endpoint	Description
`POST`	`/api/chat`	Send a message, get a response
`POST`	`/api/chat/stream`	Send a message, get streaming SSE response
`GET`	`/api/sessions`	List all active sessions
`GET`	`/api/prompts`	List available prompt strategies
`GET`	`/health`	Health check

Example Request:

curl -X POST http://localhost:8000/api/chat \
  -H "Content-Type: application/json" \
  -d '{
    "message": "What are the library timings?",
    "session_id": "user-123",
    "prompt_style": "xml_tags"
  }'

Multi-Agent RAG Endpoints

Method	Endpoint	Description
`POST`	`/ask`	Ask a question (routes through orchestrator)
`POST`	`/upload`	Upload a document (PDF, TXT, or image)
`POST`	`/ingest`	Ingest sample documents from `sample_docs/`
`GET`	`/sources`	List all ingested document sources
`GET`	`/health`	Health check

Example Request:

curl -X POST http://localhost:8001/ask \
  -H "Content-Type: application/json" \
  -d '{
    "question": "What is the attendance policy?"
  }'

🎓 Bonus Topics

🚀 Advanced Topics Covered in Workshop

Topic	Description
🔄 CI/CD	GitHub Actions or Cloud Build auto-deploy on push
📈 Autoscaling	Tuning min/max instances, concurrency, load testing
📊 Monitoring	Cloud dashboards, latency/error alerts, logging
💰 Cost Optimization	Model tier selection, token tracking, query caching
🧩 Agent Development Kit	Framework alternative to manually-built agent loops
🤏 Small Language Models	When smaller/lighter models beat large ones
📏 Evaluation	Golden datasets, LLM-as-judge, human review
🌀 Hallucination	Why it happens, mitigation beyond RAG grounding

🤝 Contributing

Contributions are welcome! If you attended the workshop and want to improve or extend the projects:

Fork the repository
Create a feature branch (git checkout -b feature/amazing-feature)
Commit your changes (git commit -m 'Add amazing feature')
Push to the branch (git push origin feature/amazing-feature)
Open a Pull Request

📬 Contact

Vaishnavi Wangalwar

Navigation Menu

Repository files navigation

From Zero to Production-Ready AI Applications

🗺️ Workflow

✨ Highlights

🏗️ Production Architecture

🧪 Hands-On Learning

🔧 Real-World Tools

🚀 Deploy Anywhere

📦 Repository Structure

🚀 Quick Start

Prerequisites

⚡ Project 1 — ChatBot Agent Setup

⚡ Project 2 — Multi-Agent RAG Setup

📦 Project 1 — ChatBot Agent

🎯 What You'll Learn

🧠 Project 2 — Multi-Agent RAG Assistant

🏗️ Architecture & Data Flow

🎯 What You'll Learn

📚 Workshop Modules

☁️ Deployment

🛠️ Tech Stack

🔌 API Reference

🎓 Bonus Topics

🤝 Contributing

📬 Contact

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
chatbot_agent		chatbot_agent
multi_agent_rag		multi_agent_rag
.gitignore		.gitignore
Closing.md		Closing.md
README.md		README.md

Latest commit

History

Folders and files

Repository files navigation

From Zero to Production-Ready AI Applications

🗺️ Workflow

✨ Highlights

🏗️ Production Architecture

🧪 Hands-On Learning

🔧 Real-World Tools

🚀 Deploy Anywhere

📦 Repository Structure

🚀 Quick Start

Prerequisites

⚡ Project 1 — ChatBot Agent Setup

⚡ Project 2 — Multi-Agent RAG Setup

📦 Project 1 — ChatBot Agent

🎯 What You'll Learn

🧠 Project 2 — Multi-Agent RAG Assistant

🏗️ Architecture & Data Flow

🎯 What You'll Learn

📚 Workshop Modules

☁️ Deployment

🛠️ Tech Stack

🔌 API Reference

🎓 Bonus Topics

🤝 Contributing

📬 Contact

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages