Meeting Minutes Internal Office Assistant: Citation and

Data Characteristics

Internal meeting minutes in the biopharmaceutical domain originate from internal meeting systems, collaboration platforms, or email archives. Update frequency aligns with meeting schedules, often several times per week or month, indicating high timeliness. Document structure typically includes consistent fields: meeting topic, time, location, attendees, agenda, discussion content, decisions, action plans, and responsible parties. Discussion content often contains extensive industry-specific terminology, drug names, experimental data, project progress, and regulatory requirements, frequently in unstructured text. Fields and units require careful handling; for example, project cycles in "weeks" or "months," and experimental data in biochemical units like "mg/kg" or "μM."

Constraints on Citation and Traceability

High timeliness of meeting minutes requires rapid indexing of new content in the knowledge base to ensure accurate and timely citations. Fixed document structures enable leveraging structured fields for key information extraction, such as quickly locating meeting resolutions via the "decisions" field. However, unstructured discussion content, especially with specialized terminology and experimental data, challenges the semantic understanding of RAG (Retrieval-Augmented Generation) systems. This necessitates more refined text segmentation and vectorization strategies to prevent misinterpretations or out-of-context citations. Citing specific fields like action plans and responsible parties requires the system to provide the original text and clearly mark the source of these key details. This allows users to quickly trace back to specific paragraphs in the original meeting minutes for accountability and follow-up.

Configuration Settings

Configuration Item	Recommended Value	Rationale
`Chunk size` (Segment Length)	`300–500 characters`	Balances the coherence of meeting discussion content with RAG retrieval efficiency. Avoids overly long or short segments that lose context or introduce noise.
`Overlap Length`	`50–100 characters`	Ensures semantic continuity at segment boundaries, improving the completeness of recalled snippets.
`Recall count` (Recall Count)	`Top 5–8 entries`	Covers potentially dispersed key information points in meeting minutes while controlling the input length for the generation model.
`Similarity threshold` (Similarity Threshold)	`0.75–0.85`	Filters out less relevant segments, focusing on meeting content highly pertinent to the user query.
`Rerank result count` (Reranked Return Count)	`Top 3 entries`	Further refines the most relevant segments from the recalled set, optimizing final citation quality.
`Max Token Limit`	`2000–3000`	Accommodates the potentially longer nature of meeting minutes, ensuring sufficient contextual information is included.

Common Pitfalls

Citation source filename does not match actual content. The citation link points to a file that does not align with the model's output. This occurs when knowledge base indexing or original file path mappings are outdated or incorrect.
Model returns incomplete or semantically biased citation snippets. Cited text is truncated mid-specialized terminology, making it incomprehensible to the user. This results from an overly aggressive segmentation strategy that does not adequately consider the integrity of biopharmaceutical terminology.
External API calls with detail: true result in failed parsing or missing citation information. The frontend page fails to display the cited knowledge base source parameters correctly. This happens when the external system's parsing logic for the FastGPT API's citation structure is incomplete or not updated to the latest version.

Verification Steps

Upload representative meeting minute documents. Check if the knowledge base index status shows "completed" and verify if the segment preview meets expectations.
For queries containing specific technical terms and decisions, observe if the model's generated answer accurately cites the relevant paragraphs in the original meeting minutes.
Simulate uploading meeting minutes at different times. Verify if the system prioritizes citing the most recently updated document content to ensure timeliness.
Use the API interface to call and inspect the citation source information returned by the detail parameter. Confirm that fields like cited content, filename, and page number are complete and parse correctly.

The values provided are common starting points and should be measured against the reader's own samples.

Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.