Contents

the contents section in pusaka allows you to store and manage various reference materials or documents these contents are indexed for retrieval by the ai assistant, improving the accuracy and reliability of responses while minimizing hallucinations overview within the contents section, users can view and filter existing reference materials (e g , pdfs, books, articles) add new content via file upload configure chunking and permission access update or delete content to keep the reference library current contents summary 1\ filtering and searching use filters for efficient lookup based on code unique identifier for each content item title descriptive name of the document author person or source responsible for the document state content status (e g , approved , pending ) buttons available apply executes filter search reset clears filters and returns full content list 2\ content table displays an overview of all uploaded contents with columns for code — unique id title — document title state — workflow status author — assigned author category — content type active/inactive chunks — segmentation count for search indexing 3\ adding new content click new content to launch the content creation form content detail fields include field description code unique identifier (e g , ocrbra , tip002 ) author select from existing authors title descriptive document title file upload file (currently supports pdf and text based formats) chunking strategy choose method for dividing the file (e g , semantic chunking & skimming chunking ) content thumbnail optional image url for cover/preview content source optional reference or citation source url permissions access restriction code for authorized viewing only buttons save submits the new content cancel discards unsaved changes 4\ chunking strategy the chunking strategy defines how uploaded content will be segmented semantic chunking (default) uses content meaning to create contextually coherent chunks semantic chunking parameter skimming chunking quickly divides content into sections based on visual or structural markers (e g , headings, bullet points) without deep semantic analysis ideal for rapid indexing of well formatted documents skimming chunking parameter advanced parameters (via ⋯ ) may include fixed length or manual segmentation options depending on platform settings 5\ permissions use the permissions dropdown to define access control via permission codes only users assigned the same permission code can view or use this content during assistant responses recommended file guidelines before uploading, prepare your documents using the following best practices use plain text or well formatted pdfs to avoid parsing issues remove all headers/footers including page numbers repeat table headers on each page to maintain clarity when chunked 6\ managing content edit click on an existing entry to update metadata or upload a revised file delete removes content from the system (irreversible) best practices use structured naming e g , med2025 guide , tips runner beg choose accurate chunking semantic chunking is best for natural language documents assign the correct author helps track source origin and improves traceability maintain access control use permission codes for confidential or internal documents keep source urls valid if referencing external data, ensure the content source remains accessible next steps after adding or editing content visit reference configuration to ensure the document is utilized in ai responses check the chunks section to verify accurate segmentation review the authors list for consistency and completeness