BotBhai 🤖💼

BotBhai is a self-service SaaS platform that allows professionals, content creators, and businesses to build, customize, and embed personalized chatbots powered by Retrieval-Augmented Generation (RAG).

By uploading resume files, product documentation, policy manuals, or FAQs, BotBhai ingests the knowledge, chunks it, and builds a dedicated context retriever. It features a unique Feedback Loop (Missing Answers) system that captures questions the chatbot couldn't answer, enabling bot owners to update the chatbot's knowledge base on the fly.

---

🌟 Key Features

• Document Management & Ingestion:
  • Supports .pdf, .txt, .md, and .csv files.
  • Automated text parsing and cleaning.
  • Smart recursive chunking with custom overlap limits for high-fidelity vector searches.
• PII (Personally Identifiable Information) Detection:
  • Scans documents on upload using regex and OpenRouter AI validation.
  • Detects emails, phone numbers, physical addresses, and government ID patterns.
  • Allows owners to choose between auto-redaction ([REDACTED]) or storing the original.
• Embeddable Widget:
  • Generates a simple, zero-dependency HTML script snippet.
  • Injects a floating, responsive chat assistant onto any website or blog.
• Missing Answers Log (Feedback Loop):
  • Logs questions from visitors that received low-confidence or "I don't know" answers.
  • Groups and tallies identical missing queries.
  • Provides a one-click "Add Data" feature to instantly append missing information.
• Custom Persona & Configurations:
  • Customize bot name, system instructions, and response tone (Professional, Friendly, Humorous).
• Sources & Attributions:
  • Displays collapsible references showing the exact document chunks and text snippets used to formulate answers.

---

🛠️ Technology Stack

Layer	Technology	Role
Frontend	React, Next.js (App Router), Tailwind CSS	Core UI and Dashboard
Database	Google Cloud Firestore	Metadata, session state, and config storage
Authentication	Firebase Authentication	Google Sign-in + Email/Password
Storage	In-memory ingestion (originals not persisted)	Files are parsed → embedded on upload
Vector DB	Pinecone	Dense vector similarity retrieval (768 dims, cosine)
Embeddings Model	Jina jina-embeddings-v3	Generates 768-dimension text embeddings (Matryoshka)
Chat LLM	Groq llama-3.1-8b-instant	Context-aware answer generation

---

📁 Project Structure

botbhai/
├── app/
│   ├── (public)/              # Public landing page and static files
│   │   ├── page.tsx           # Landing page
│   │   └── embed.js           # Static widget script for floating bubble
│   ├── (dashboard)/           # Protected dashboard panel
│   │   ├── layout.tsx         # Dashboard shell & sidebar layout
│   │   ├── page.tsx           # /dashboard - Documents and uploading
│   │   ├── test-chat/         # /dashboard/test-chat - Sandbox testing
│   │   ├── missing/           # /dashboard/missing - Unanswered questions log
│   │   ├── embed-config/      # /dashboard/embed-config - Embed code copier
│   │   └── analytics/         # /dashboard/analytics - Usage statistics
│   └── api/                   # API routes (Controllers)
│       ├── documents/         # Upload/list/delete knowledge docs
│       ├── chat/              # Public chat conversation endpoint
│       ├── missing/           # Read/resolve unanswered questions
│       ├── bot/               # Configuration settings
│       └── analytics/         # Usage stat aggregations
├── components/                # Reusable React components (Views)
│   ├── ui/                    # Design system primitives (Button, Card, Modal, etc.)
│   ├── dashboard/             # Page-specific feature layouts
│   └── chat/                  # Public floating bubble widgets
├── lib/                       # Business logic layer (Models & Services)
│   ├── db/                    # Firestore CRUD adapters
│   ├── vector/                # Pinecone client & embedding integrations
│   ├── ai/                    # OpenRouter LLM interface (Chat, PII, Rewriter)
│   └── chunking/              # Recursive text splitting algorithms
├── types/                     # Global TypeScript interfaces
└── scripts/                   # CLI maintenance & validation scripts

---

⚙️ Environment Configuration

Create a .env.local file in the root directory (based on .env.local.example):

# Firebase Client SDK Credentials (Publicly exposed)
NEXT_PUBLIC_FIREBASE_API_KEY=your_firebase_api_key
NEXT_PUBLIC_FIREBASE_AUTH_DOMAIN=your_project.firebaseapp.com
NEXT_PUBLIC_FIREBASE_PROJECT_ID=your_project_id
NEXT_PUBLIC_FIREBASE_STORAGE_BUCKET=your_project.appspot.com
NEXT_PUBLIC_FIREBASE_MESSAGING_SENDER_ID=your_sender_id
NEXT_PUBLIC_FIREBASE_APP_ID=your_app_id

# Firebase Admin SDK Credentials (Server-only)
# Format: Single-line stringified JSON of service account private key
FIREBASE_ADMIN_SDK_JSON={"type":"service_account","project_id":"...","private_key":"..."}

# Pinecone Integration (index: 768 dims, cosine)
PINECONE_API_KEY=your_pinecone_api_key
PINECONE_INDEX_NAME=botbhai-mvp

# Jina embeddings
JINA_API_KEY=your_jina_api_key
JINA_EMBED_MODEL=jina-embeddings-v3
JINA_EMBED_DIM=768

# Groq LLM
GROQ_API_KEY=your_groq_api_key
GROQ_MODEL=llama-3.1-8b-instant

# Web Client URL
NEXT_PUBLIC_APP_URL=http://localhost:3000

---

🚀 Installation & Local Development

1. Prerequisites

Ensure you have Node.js v18+ installed. Set up your Pinecone index with 768 dimensions and cosine distance metric.

2. Install Dependencies

npm install

3. Firestore Indexes Setup

Make sure the following composite indexes are deployed in your Firebase Firestore database (configured in firestore.indexes.json):

Collection ID	Field 1	Field 2	Field 3	Query Scope
documents	userId (Ascending)	uploadedAt (Descending)	—	Collection
documents	botId (Ascending)	uploadedAt (Descending)	—	Collection
missing_entries	botId (Ascending)	status (Ascending)	lastSeen (Descending)	Collection
missing_entries	botId (Ascending)	status (Ascending)	timesAsked (Descending)	Collection

To deploy Firestore indexes via Firebase CLI:

firebase deploy --only firestore:indexes

4. Running the Development Server

npm run dev

Open http://localhost:3000 to view the client.

5. Running Integration Tests

Verify Jina, Groq, and Pinecone connections by executing the integration test suite:

npm run test:integrations

---

🔒 Security & Limits

• PII Check: Optional scanning verifies context prior to DB save.
• Daily Sessions: Limited to 200 sessions per bot.
• Concurrency: In-memory tracking restricts bots to 15 concurrent sessions.
• Retention: Chat logs auto-purge after 30 days, and unanswered questions are retained for 90 days.