A vectorless RAG (Retrieval Augmented Generation) implementation that uses tools like ripgrep and semantic document analysis instead of traditional vector databases.

• PDF Document Processing: Upload and automatically extract text, generate summaries, tags, and references for each page
• AI-Powered Query Processing: Use GPT-4 with tool-calling capabilities to search through documents intelligently
• Semantic Search: Find relevant content using natural language queries
• Reference Following: Automatically discover and follow cross-references between document pages
• Type-Safe API: Built with TRPC for end-to-end type safety

• Runtime: Bun
• Framework: Next.js 14 with App Router
• Frontend: React with Tailwind CSS
• Backend: Next.js API Routes with TRPC
• AI: Vercel AI SDK with OpenAI GPT-4
• PDF Processing: pdf-parse

1. Install dependencies:

   bun install

2. Environment Setup: Copy .env.example to .env.local and add your OpenAI API key:

   cp .env.example .env.local

   Edit .env.local:

   OPENAI_API_KEY=your_openai_api_key_here
   

3. Development:

   bun run dev

4. Production:

   bun run build
   bun run start

1. Upload: User uploads a PDF document (100+ pages)
2. Processing: Backend extracts text from each page and creates DocumentPage structures
3. Analysis: AI generates summaries, tags, and extracts references for each page
4. Query: User asks questions about the document
5. Search: AI agent uses tools to gather relevant context across multiple pages
6. Answer: System provides comprehensive answers with source references

Each page contains:

• pageNbr: PDF page number
• pageContentNbr: Page number shown in document (optional)
• content: Extracted text content
• summary: AI-generated 1-5 sentence summary
• tags: Semantic topic tags for retrieval
• references: Section numbers, chapter references, etc.

The query agent has access to:

• search_pages: Find pages containing keywords
• get_page_content: Retrieve full page content
• search_references: Find pages referencing specific sections
• get_related_pages: Find semantically related pages

• uploadDocument: Process and store PDF documents
• queryDocument: Query documents with natural language
• getDocuments: List all uploaded documents

Perfect for:

• Legal documents with complex cross-references
• Technical manuals with section references
• Academic papers with citations
• Regulatory documents with circular references

The system creates an Abstract Syntax Tree-like structure for documents, enabling:

• Semantic understanding without vectors
• Tool-based search and retrieval
• Cross-reference following
• Multi-step reasoning across document sections