Knowledge Base
GAv2.8.0Upload PDFs (menus, FAQs, policies) so Beacon answers customers from your real information, not guesswork.
updated 5 days ago
Live previewinteractive
Indexing pipeline
Haystack · Qdrant · Neo4j
01
Upload PDF100%
S3 · max 50 MB
02
Gemini semantic chunking100%
200–400 words · or per-FAQ
03
Embed · dense + sparse100%
Cohere multilingual-v3 + SPLADE++
04
Write to Qdrant100%
filter key: knowledge_base_id
Query · “what's your refund window?”
hybrid → rerank → graph → rerank
source / chunksparsedensererank
refund-policy-v4.pdf
chk_8f2e
0.81
0.93
0.97
shipping-faq.pdf
chk_1a0c
0.62
0.79
0.84
returns-process.pdf
chk_4de2
0.45
0.71
0.72
A simple view of what Beacon has learned and what it cites in answers.
§01 · Mechanics
How it works
3 steps · end-to-end
- 01Upload your docs
Menus, pricing, policies, and any business handbook.
- 02Beacon learns them
Beacon indexes the document so it can answer quickly and consistently.
- 03Answers stay on-brand
Beacon replies using your wording and your rules, with fewer hallucinations.
§02 · In the wild
Where teams actually use it
Restaurantscase 01
Menu + allergy answers
Answer ingredients, add-ons, pricing, and opening hours correctly.
Retailcase 02
Returns + warranty policies
Consistent answers even during busy hours and holidays.
Hotelscase 03
Check-in, parking, amenities
Fewer repetitive questions for staff.