Knowledge Sources and Quality Upgrades for Sparkforge Conversation AI

Sparkforge Conversation AI now ingests tables, rich-text, and file uploads, and a re-ranking layer picks the single best snippet before the model answers. Source attribution appears on every response.

What's New

• Higher accuracy: automatic re-ranking surfaces the most relevant chunk before generation.
• Faster rollout: add or replace data sources without retraining from scratch.
• Flexible training: spreadsheets, rich-text, PDFs, and more.
• Source visibility: every response includes clickable source links in the Response Info drawer.

Smarter Retrieval with Re-Ranking

A ranking layer sits between vector search and answer generation. After initial retrieval, the re-ranker scores each chunk for semantic closeness to the user question and forwards only the top passages to the LLM. The result: fewer hallucinations, tighter replies.

Expanded Data Type Support

• Paste or upload Tables (CSV/TSV) so the AI can reference structured values like pricing grids.
• Add Rich-Text blocks with headings, bold, and lists directly in the editor.
• Upload Files (PDF, DOCX, PPT, TXT). The system auto-chunks for search.

Source Attribution

Every AI response inside Conversations shows a Response Info icon. Open the drawer to see:

• The exact knowledge chunk(s) used.
• File or URL name, FAQ label, and timestamp.
• Quick-edit options to correct or replace the source on the fly.

Setup

Step 1: Create or Open a Knowledge Base

1. Sign in to your sub-account.
2. Open the AI Agents tab from the left menu.
3. Click the Knowledge Base tab.
4. Click + Create Knowledge Base or open an existing one.

Step 2: Add Sources

Pick the source type — table, rich-text block, or file upload — and the system auto-chunks and indexes the content.

Step 3: Test in Conversations

Trigger a test message, then open the Response Info drawer to confirm the bot pulled from the right source.