Dokko, our intelligent assistant designed to simplify complex knowledge retrieval and chatbot creation, continues to evolve. For those unfamiliar, Dokko is an AI-powered platform that combines advanced language models, document understanding, and Retrieval-Augmented Generation (RAG) to enable seamless deployment of custom chatbots. Whether you’re building a customer support bot or an internal knowledge assistant, Dokko helps you get started in minutes—delivering accurate, contextual answers across various formats and data sources.

We’re excited to share what we’ve been working on this past quarter. Here’s a detailed look at the newest improvements:

Multimodal Support: Images and Tables in PDFs

Dokko now offers full multimodal capabilities! Leveraging powerful vision models, Dokko can extract and understand images and tables from PDF documents, ensuring these visual elements are integrated into answer generation. This enhancement is particularly useful for technical documentation, scientific papers, and visually rich materials.

Direct Linking to PDF References

Users can now click directly into a PDF file to view the exact page where a relevant text chunk was found. This new feature enhances transparency and allows for deeper verification and context exploration.

Intelligent Query Classification: Chit-Chat vs. RAG

Not every conversation needs document retrieval. Dokko now includes preprocessing logic to identify whether a query is casual “chit-chat” or requires RAG-based retrieval. This makes interactions more fluid and efficient by avoiding unnecessary document searches.

Smarter RAG: Translation and Rich Metadata

To boost retrieval accuracy, especially for multilingual content, Dokko can now automatically translate documents to English during preprocessing. On top of that, we’ve enriched our embedding vectors with structured metadata:

[Excerpt from Document]
Excerpt title: {title}
file_path: {file_path}
questions_this_excerpt_can_answer:
{questions_txt}
section_summary: {summary}
keywords: {keywords}
Excerpt:
-----
{content}
-----

This structured format enhances both semantic search and metadata-based retrieval, improving response quality and contextual awareness.

Enhanced Full-Text Search with Metadata Fields

Full-text search, a part of our hybrid approach and a valuable addition to vector search, just got smarter. Each document chunk now stores the following metadata:

• page_number: Page where the chunk appears (if available)
• tags: List of associated content tags
• source: File download location
• summary: Brief summary of the chunk
• keywords: Extracted terms from the text
• file_name: Original file name
• questions: Up to two questions the chunk can answer
• chunk_title: Title or heading of the chunk

These enhancements make searches more targeted and informative.

Web Crawler Integration

No more manual uploads. With the new web crawler functionality, Dokko can now automatically download and process content from specified URLs. Keeping your knowledge base fresh and comprehensive has never been easier.

Foundations for AI Agents

We’ve begun laying the foundation for AI agents within Dokko. Although in the early stages, this initiative will enable Dokko to perform tasks autonomously, delegate actions, and interact with external systems—expanding its capabilities well beyond simple Q&A.

We’re thrilled about Dokko’s progress and remain committed to making it the most powerful and user-friendly AI assistant platform available.

Stay tuned—there’s much more to come and visit dokko.ai to explore the platform or start building your assistant today!