Hannah: AI-Powered Knowledge Assistant
Transforming customer support from hours of searching to instant, accurate answers with AI
The Challenge
Customer support teams face a constant challenge: answering the same questions over and over while ensuring accuracy and consistency. Support agents spend valuable time:
- •Searching through multiple documentation pages to find the right answer
- •Manually copying and customizing responses from knowledge bases
- •Worrying about giving outdated or incorrect information
- •Training new team members on where to find information
The result? Longer response times, inconsistent answers, and support teams stretched thin. One team we spoke with estimated spending 3-5 minutes per ticket just searching for the right information to share with customers.
The Solution
We built Hannah—an AI assistant that transforms any company's website and documentation into an intelligent, searchable knowledge base. Support agents can ask questions in plain English and get instant, accurate answers with source citations.
Key Features
Instant Answers
Get responses in under 2 seconds with zero hallucination—all answers are grounded in actual company content.
Source Citations
Every answer includes clickable links to the exact pages where the information came from for easy verification.
Smart Web Crawling
Automatically crawls and indexes website content, extracting only valuable information while skipping ads and navigation.
Always Up-to-Date
Re-crawl websites on demand to keep the knowledge base current as documentation changes.
How It Works
Website is crawled: Hannah automatically scans the company website, extracting content from documentation pages, FAQs, product pages, and blog posts.
Content is indexed: The content is broken into searchable chunks and stored with AI embeddings for fast, semantic search.
Support agent asks a question: Instead of searching manually, the agent types a natural language question into Hannah.
AI retrieves and responds: Hannah finds the most relevant content and generates a clear, accurate answer with links to source pages.
The Results
By eliminating the manual search process, Hannah transformed how support teams work:
- ✓Support agents spend less time searching and more time helping customers
- ✓New team members get up to speed in days instead of weeks
- ✓Customers receive faster, more accurate responses
- ✓Documentation stays the single source of truth with automatic updates
Technical Implementation
Technical Implementation
Architecture
Frontend
- • Next.js 14 with TypeScript for type safety
- • TailwindCSS for responsive design
- • Framer Motion for smooth animations
- • Deployed on Cloudflare Pages (auto-deploy from GitHub)
Backend
- • FastAPI (Python) for high-performance API endpoints
- • SQLite database for edge compatibility
- • OpenAI embeddings (text-embedding-ada-002) for semantic search
- • GPT-4o-mini for response generation
- • Deployed on Railway.app with automatic scaling
RAG Pipeline
- • Smart web crawler with content filtering
- • Chunks content into 512-token segments
- • Cosine similarity search for retrieval (no external vector DB needed)
- • Binary quantization for fast similarity matching
- • Quality scoring and deduplication
Key Design Decisions
- • Zero hallucination guarantee: Strict system prompts ensure AI only uses retrieved content
- • Source transparency: Every answer shows exactly where information came from
- • Cost optimization: SQLite instead of managed vector database reduces infrastructure costs
- • Edge deployment: Fast global performance with Cloudflare Pages CDN
Want to build something similar?
Let's talk about how AI can transform your team's workflow. Every business has repetitive tasks—we help you automate them.
Start Your Project