How Leksa Works

A short guide to the product: what it does, how you manage content and lawyers, and how we keep improving answers.

Start overview

The big picture

When someone uses Leksa, this is the flow—without going into technical detail.

User asks a question → We find relevant passages in your documents → AI answers in plain language (with citations) → When appropriate, we suggest talking to a lawyer
1
Question

The user types a question in the chat—in English or Swahili. They can pick a topic (e.g. property, employment) so answers focus on that area.

2
Search your knowledge

The system searches only the documents you’ve added and trained. It picks the most relevant passages to answer the question.

3
Answer

The AI writes an answer based on those passages, in simple language. It can cite sections or sources so the user knows where the information comes from.

4
Referral to a lawyer

When the question needs professional advice or the system isn’t confident enough, it suggests consulting a lawyer and can link to your lawyer directory.

What’s in the product

Main features your firm and your clients will use.

Chat

Users get a simple chat interface. They can ask follow-up questions; the system keeps the conversation in mind and can narrow answers to a chosen topic (category).

Topics (categories)

You define legal topics (e.g. Business law, Property, Family). Each topic can have an icon and description. Users see these on the chat home screen and can choose one so answers are limited to documents in that category.

Document management

You upload PDFs and documents, assign them to a category, and “train” the system. Once trained, those documents become the knowledge base the AI uses to answer questions.

Lawyer directory

A public directory of lawyers. You manage categories (e.g. by practice area) and list lawyers with their details. Users can browse by category when the AI suggests talking to a lawyer.

Handoff to a lawyer

When the AI decides the user should speak to a lawyer, the answer can include a clear suggestion and a button that takes them to the lawyer directory—optionally filtered by the relevant category.

Settings

You can turn the lawyer referral on or off, change the label (e.g. “Lawyer” or “Advocate”), enable a credits system for questions, and manage API and system options from the admin.

Document management

How the content the AI uses is controlled by you.

Adding and organizing documents
  • In the admin you upload documents (PDF, Word, or text). You give each one a title and optionally assign it to a category (the same topics users see on the chat).
  • Categories help you organize by practice area. When a user selects a topic in chat, the system only searches documents in that category—so answers stay on topic and can be faster.
  • You can activate or deactivate documents. Only active documents are used when answering questions.
Training (indexing)
  • After upload, each document must be trained. Training means the system processes the text and builds a searchable index. Until a document is trained, it won’t be used in answers.
  • You trigger training from the document list. When it’s done, the document status shows as indexed and the AI can use it.
  • If you update a document or replace the file, you can run retrain so the index reflects the latest content.

Lawyers

How the lawyer directory fits in and how you maintain it.

  • Categories: You create lawyer categories (e.g. Property, Family law, Commercial). These can align with your document topics so that when the AI suggests a lawyer, it can point to the right category.
  • Listing lawyers: For each lawyer you add a name, contact details, and category. The public directory page shows lawyers by category so users can find someone relevant.
  • Visibility: The lawyer directory and the “talk to a lawyer” suggestion in chat can be turned on or off in settings. You can also change the label (e.g. “Lawyer” or “Advocate”) so it matches your branding.

How questions move from AI to lawyers

When and how we suggest talking to a human.

  • Not every question should be answered only by the AI. When the system detects that the user needs professional advice, or when it’s not confident that the documents fully support an answer, it can add a referral: a short suggestion to consult a lawyer.
  • The answer the user sees will still contain the best information from your documents. At the end, we can show a clear line like “For advice tailored to your situation, consider speaking to a lawyer” and a button.
  • That button takes the user to the lawyer directory. When possible, we link to the directory filtered by the category that matches the question (e.g. property), so they see lawyers in the right practice area.
  • So: the AI handles first-line information; when it’s appropriate, we hand the user to your lawyers in a structured way, without exposing any internal mechanics.

How we train the system

In plain terms—no technical jargon.

  • You provide the content. The only knowledge the AI uses for legal answers is the documents you upload. There is no generic legal database; it’s your materials, organized by category.
  • Training = making it searchable. When you “train” a document, the system breaks the text into manageable pieces and builds an index. When a user asks a question, we search this index for the most relevant pieces and then the AI writes an answer based on them.
  • Categories focus answers. By assigning documents to categories and letting users choose a topic, we limit search to that topic. That keeps answers relevant and can improve speed.
  • Quality of answers depends on the quality and coverage of the documents you add. The more accurate and up to date your content, the better the answers. You can retrain when you update a document.

What we monitor

So you can see how the system is used and where to improve.

Query logs

We keep a record of questions asked, so you can see what users are interested in and whether certain topics come up often.

Analytics

High-level stats on how many queries were answered, how often we suggested a lawyer, and how the pipeline performed over time.

Feedback

Where enabled, users can give simple feedback (e.g. thumbs up/down) on answers. That helps spot answers that were helpful or need work.

RAG runs

For each question we can see which documents and passages were used and whether the system decided to refer to a lawyer. Useful for debugging and improving content.

How we improve it

Ongoing steps to make answers better and more reliable.

  • Feedback and logs: We use query logs and any user feedback to see which questions are asked often and which answers might need improvement. That can guide what new documents to add or which categories to expand.
  • Content updates: When laws or internal policies change, you add or update documents and retrain. The system then uses the new content for future answers.
  • Validation: The system checks that answers are grounded in the retrieved passages. When the match isn’t strong enough, it can try again with different passages or suggest a lawyer instead of guessing.
  • Referral tuning: We can adjust when the AI suggests a lawyer (e.g. for sensitive topics or when confidence is low) so that users get a clear path to human advice when they need it.