Knowledge Base — Train Your Bot

Knowledge Base — Train Your Bot

Your bot's intelligence comes entirely from your knowledge base. The more relevant content you add, the better it answers customer questions. This guide covers all three ways to train your bot and tips for getting the best results.

---

How It Works Behind the Scenes

When you add content, Botaura breaks it into small chunks (800–1000 tokens each), converts each chunk into a mathematical representation (embedding), and stores it in a vector database. When a customer asks a question, the bot:

1. Converts their question into the same mathematical space
2. Finds the most similar chunks from your knowledge base
3. Reads those chunks and generates a natural, conversational answer

This is called RAG (Retrieval-Augmented Generation). The bot never makes up answers — it only uses your content.

---

Method 1: Upload Documents

Go to Dashboard → Knowledge Base → File Uploads.

Supported formats:

• PDF — product catalogs, brochures, price lists
• DOCX — Word documents with business info, policies
• TXT — plain text files
• MD — Markdown files

Tips for Better Results

• Use clean, well-structured documents — headings, bullet points, clear sections
• Remove headers/footers, page numbers, and decorative text that adds noise
• One focused topic per document works better than one massive file
• Include your most frequently asked questions in a dedicated FAQ document

Limits

Plan	Max Files	Max Size per File	Knowledge Chunks
Free	5	2 MB	100
Pro	Unlimited	50 MB	5,000
Business	Unlimited	50 MB	5,000
Growth	Unlimited	50 MB	5,000

---

Method 2: Crawl Your Website (Pro+)

Go to Dashboard → Knowledge Base → Website Crawling.

Enter your website URL and click Start Crawl. Botaura will:

1. Visit your homepage and follow internal links
2. Extract text content from each page (removing navigation, sidebars, footers)
3. Detect product pages and import product details (name, price, image, description)
4. Break the content into chunks and add to your knowledge base

What Gets Crawled

• All text content on your pages
• Product information (detected via price patterns, "Add to Cart" buttons, structured data)
• Sitemap pages (if available)

What Doesn't Get Crawled

• Pages blocked by robots.txt
• Login-protected pages
• External links (other websites)
• Images and videos (only text is extracted)

Auto-Recrawl (Pro+)

Enable automatic recrawling to keep your knowledge base up to date. When your website content changes, Botaura will re-crawl and update the knowledge base automatically. Configure the schedule in Settings → Auto-Recrawl.

---

Method 3: Manual Entries

Go to Dashboard → Knowledge Base → Manual Entry.

Two formats available:

Free Text

Paste any content — about us page, return policy, service descriptions, operating hours. Good for information that's not on your website or not in a document.

Q&A Pairs

Structured question-and-answer format. Perfect for FAQs:

• Q: What are your delivery charges?
• A: We deliver free within Lahore. Outside Lahore, delivery is Rs 200.

Q&A pairs give the best results because the bot can directly match customer questions to your answers.

---

Testing Your Knowledge Base

After adding content, use the Test Search feature to verify:

1. Type a question your customers might ask
2. See which chunks the bot finds and their confidence scores
3. High confidence (0.7+) — the bot found a great match and will answer confidently
4. Medium confidence (0.4–0.7) — the bot found something relevant but might supplement with general knowledge
5. Low confidence (below 0.4) — the bot didn't find a good match and will offer to connect the customer with your team

Improving Results

If the bot isn't finding the right answers:

• Add more content on that specific topic
• Create Q&A pairs for common questions the bot misses
• Use the exact phrases your customers use (including Urdu/Hinglish terms)
• Break long documents into focused sections

---

Multilingual Content

Your knowledge base can be in English — the bot handles language translation automatically. When a customer asks in Roman Urdu, the bot translates their question to English for searching, finds the answer, then replies in Roman Urdu.

However, adding some content in Urdu/Hinglish can improve results for very specific local terms that don't translate well (product names, local phrases, brand names in Urdu).

---

Managing Your Knowledge Base

• Delete outdated content — remove files or entries that are no longer accurate
• Re-crawl after website updates — if you update your website, re-crawl to refresh the knowledge base
• Monitor the chunks count — check your usage in Dashboard → Pricing to see how many chunks you've used
• Quality over quantity — 50 well-written Q&A pairs beat 500 pages of irrelevant content