AI Learning & RAG (Voyage Embeddings)

Overview

BlueClerk's AI chat assistant learns from every conversation and past job specific to YOUR company. The more you use it, the smarter it gets - remembering customer preferences, common services, pricing patterns, and how your team refers to things. This is powered by RAG (Retrieval Augmented Generation) using Voyage AI embeddings, recommended by Anthropic.

How AI Learning Works

Company-Specific Knowledge

Every time you chat with the AI assistant:

• Question and answer pairs are saved in your company's knowledge base
• Past interactions inform future responses - the AI remembers what worked before
• Your business context is learned - customer names, common services, pricing patterns
• Tools used are tracked - so the AI knows which actions are most helpful

What Gets Remembered

The AI learns from:

• Past jobs - Similar work you've done before, including actual costs
• Chat history - Previous conversations and successful resolutions
• Customer preferences - How specific customers like things done
• Pricing patterns - What you typically charge for different services
• Common services - Work you do frequently
• Your terminology - How your team refers to things

RAG Context in Action

When you ask the AI a question, it automatically searches:

1. Similar past jobs from your company history (if you have pgvector enabled)
2. Past chat learnings scoped to your company - Q&A pairs, successful workflows, and tool usage
3. Platform knowledge - Built-in understanding of how BlueClerk works

The AI retrieves the most relevant context (above 35% similarity) and uses it to inform responses - ensuring answers are tailored to YOUR business, not generic.

What Powers This

Voyage AI Embeddings

BlueClerk uses Voyage AI's voyage-3 model for embeddings:

• 1024 dimensions - Optimized for retrieval with Claude AI
• Anthropic's recommended partner - Works seamlessly with Claude
• No OpenAI dependency - Independent, reliable embeddings

Automatic Background Processing

When you complete a chat interaction:

• Q&A pairs are embedded automatically in the background
• No action needed - learning happens behind the scenes
• Short/error exchanges are skipped - only meaningful interactions are saved
• Company-scoped storage - your learnings stay private to your business

Benefits

For Contractors

• Faster estimates - AI remembers similar jobs you've done
• Consistent pricing - Learns your typical rates for services
• Customer context - Recalls preferences and past work for each customer
• Less repetition - No need to re-explain your business each time

For Your Team

• New team members get up to speed faster - AI knows your processes
• Consistent responses - Everyone gets answers based on your actual work history
• Knowledge retention - Past learnings don't leave when employees do

Privacy & Security

• Company-scoped - Your learnings are never shared with other companies
• Secure storage - Embeddings stored in your company's database partition
• No PII leakage - Only meaningful work context is saved, not sensitive customer data

Questions

Q: Can I see what the AI has learned about my business? A: Currently, learned context is used automatically in chat responses. A dedicated RAG knowledge viewer is planned for a future release.

Q: How long does it take for the AI to learn? A: Learning is immediate - every chat interaction is embedded within seconds and available for the next conversation.

Q: Can I delete learned context? A: Not yet - this feature is planned. For now, learned context is permanent and company-scoped.

Q: Does this cost extra AI credits? A: No - embedding chat interactions happens in the background at no cost to you. Only active chat messages consume credits.