Search

FAQ - AI

Article author
Learning Center Mekari
  • Updated

FAQ

  1.  What AI features does Mekari eSign use?

Mekari eSign uses four AI features.

  • Document QnA answers your questions about a contract. It also summarizes sections and checks for risks in the document.
  • Auto Tagging classifies uploaded documents into categories automatically.
  • Auto Annotate Sign detects signatures, initials, and stamps in a document.
  • Document Context Extractor pulls predefined data fields from uploaded documents. These fields include parties, dates, and contract values.
  1. What languages do these features support?

These features support documents in English and Indonesian. The risk check feature uses bilingual search terms. For example, it searches for "penalty" and "denda", or "indemnity" and "ganti rugi". The Auto Tagging model was trained on Indonesian document types such as Surat Kuasa and Addendum.

  1. Does the AI send my documents to external servers?

No. All AI models run on Mekari internal infrastructure. The system does not send your document data to external services.

  1. Does Mekari use my documents to train its AI models?

No. The system uses your documents only for inference. Inference means the system reads your document to give answers and results. No customer document data is used for model training or fine-tuning. There is no cross-customer data exposure.

  1. Where do the AI models run?

All AI models run on Mekari internal servers. There is no external API dependency. Each feature uses a different model, but all models are hosted internally.

  1.  Can I rely on AI results for legal decisions?

No. All AI features in eSign give informational support only. They are not a substitute for legal advice. All answers and results come from document content and model knowledge. You must always do your own legal review. A disclaimer is shown with all AI responses in the eSign interface.

  1.  What is Document QnA?

Document QnA is an AI assistant inside eSign. It reads your uploaded contract and answers your questions. It can also summarize sections and extract key information. All answers come only from the content of your document. This feature is not a substitute for legal review.

  1.  What questions can I ask?

You can ask factual questions about the contract content. You can request section summaries. You can ask the system to extract key information. Examples include parties, dates, and payment terms.

  1. What can Document QnA not do?

Document QnA has the following limitations:

  • It does not give legal advice.
  • It does not compare different documents.
  • It does not do full-document summaries
  • It does not show highlighted references in the document.
  • It does not check legal compliance or enforceability.
  • It does not do jurisdiction-specific reasoning.
  1. How does Document QnA work?

When you upload a PDF, the system extracts the text. It divides the text into small chunks. It stores these chunks as numerical vectors.When you ask a question, the system finds the most relevant chunks. It sends them to the AI model. The AI model makes an answer only from those chunks.

  1. What document formats and sizes does it support?

Document QnA supports PDF files. Documents can be 1 to 100 pages. The typical document is 20 to 30 pages.

  1. How fast are the answers?

The target response time is 15 seconds or less per question. [INTERNAL ONLY] The system supports approximately 10 requests per second.

  1. What are Risk and Inconsistency Checks?

This feature extends Document QnA. It scans your contract and finds three types of problems:

  • Contradictions — clauses that conflict with each other.
  • Missing elements — absent items such as signatures, dates, or appendices.
  • Risky terms — unusual or high-risk clauses such as unlimited liability or auto-renewal.
  • This feature is not a substitute for legal review. It is a support tool.
  1. How does the risk check work?

The system runs a risk analysis when you upload a document. It divides the document into overlapping page windows. It runs an AI analysis on each window. Then it does a final reconciliation pass. This pass merges the results and removes duplicates. The system stores these findings for fast retrieval. When you ask a risk question, the system loads the stored findings. The AI model makes an answer from these findings only.

  1. What does a risk finding look like?

Each finding has five parts:

1. Type — contradiction, missing element, or risky term.

2. Severity — high, medium, or low.

3. Evidence — a quoted section from the document.

4. Location — page number, start line, and end line.

5. Explanation — a description of the problem.

  1. Can the risk check find all problems in my contract?

No. The risk check uses AI analysis on page windows. It can miss some problems. For example, it can miss a contradiction if the two conflicting clauses are far apart and do not appear in the same analysis window. The quality also depends on OCR accuracy for scanned documents. This feature helps you review faster. It does not replace a full legal review.

  1. What is Auto Tagging?

Auto Tagging classifies your uploaded document into a category automatically. For example, it can identify a document as an employment contract, an invoice, an NDA, or a meeting minutes.This feature is not a substitute for manual review of the document category.

  1. How does Auto Tagging work?

The system extracts text from your PDF with OCR. It converts the text into a numerical representation.A logistic regression classifier predicts the document category. If the classifier has low confidence, it labels the document as "Others." The "Others" label is a fallback. The AI model does not learn the "Others" category. The backend system handles "Others" separately.

  1. What happens if Auto Tagging gets the category wrong?

You can correct the category manually. The system records your correction. It uses these corrections in the next quarterly retraining cycle. Your feedback directly improves the model over time.

  1. What is Auto Annotate Sign?

Auto Annotate Sign detects the positions of signatures, initials, and stamps in your uploaded PDF. It places annotation markers at the correct locations automatically. It also assigns each detected area to the correct signer. This feature is not a substitute for manual verification of signature placement.

  1. How does Auto Annotate Sign work?

The system converts your PDF pages into images.

  1. What is the Document Context Extractor?

The Document Context Extractor pulls structured information from your uploaded documents automatically. It identifies specific data fields. Examples include parties, contract value, effective date, end date, and registration number. The fields change with the document type. This feature is not a substitute for manual verification of extracted data.

  1. How does the Document Context Extractor work?

The system receives your PDF and extracts the text. It checks the document type. It builds a custom AI prompt based on that type.

If the document is too long, the system splits the text into smaller parts. It processes each part separately and merges the results.

The system processes documents in the background. It can send results to a callback URL when processing is complete.

When the AI model cannot find a field in the document, it returns "NOT INFORMED" for that field. It does not guess or make up data.

  1. What document types does it support?

The system supports multiple document types. Current supported types include:

  • Employment Contract (fields: job title, compensation, work hours, probation period, termination clause, and others)
  • Addendum
  • Internal Memos
  • Termination Letters
  • Power of Attorney (Surat Kuasa)
  • Generic contracts (fields: parties, background, contract value, effective date, end date,registration number)