The team at Daphnis Labs redefined what’s possible for Urbanface. They delivered a bespoke, animation-heavy website that remains incredibly quick and functional. The…
Data pipelinesbuilt forretrieval.
Handle updates and deletions across documents, records and vector stores.
01Sources
02Parse
03Chunk
04Embed
05Index
06Query
What Data Can We Prepare?
Explore servicesDocument Collections
Business Records
Product Catalogues
Knowledge Updates
Restricted Collections
Index Migrations
What You Actually Get
- Source Connectors
- Parsing Jobs
- Chunking Configuration
- Embedding Jobs
- Index Schema
- Sync Checkpoints
- Data Quality Report
- Operations Runbook
- Source Connectors
- Parsing Jobs
- Chunking Configuration
- Embedding Jobs
- Index Schema
- Sync Checkpoints
- Data Quality Report
- Operations Runbook
A policy changes. Search catches up.
Read this example
The travel allowance guide has a new version.
- Revision received. Travel guide v4 replaces v3. Its access group remains Finance. Change detected.
- Only the change is indexed. New passages are embedded. The superseded version is excluded. Updating searchable content.
- The current source appears. Search returns the v4 allowance passage with its document link. Version + permissions retained.
A source revision updates the index without losing version or access metadata.
Scripted illustration using sample information, not a client case study or a live system.
What Does It Take to Build?
Get a custom estimatePilot
- One source collection
- Sample ingestion
- Initial retrieval checks
- Recommended
Production
- Scheduled or event-based sync
- Multiple source formats
- Live query workload
Enterprise
- Several collections or tenants
- Large-scale backfills
- Migration requirements
- Ongoing data operations
- Founded in
- 2013
- Projects delivered
- 550+
- Client countries
- 43+
- Global offices
- 3
FAQs
Do we always need a vector database?
No. Keyword search, structured queries or a hybrid approach may fit the workload. We compare retrieval needs, filters and operating requirements before selecting a store.
What happens when a source is deleted?
Deletion handling should remove or invalidate the corresponding indexed content and track completion. Retention and backup behavior also need to follow the agreed policy.
Can we change embedding models later?
Yes, but the change may require re-embedding and a new index. We plan compatibility checks, comparison queries and a cutover process rather than mixing incompatible vectors.
How do permissions reach the index?
The pipeline needs stable source identifiers and access metadata. The query layer must apply the relevant filters; indexing a permission field alone does not enforce access.
Can you process scanned or poorly structured files?
We assess extraction quality on representative samples. OCR or specialised parsers may be needed, and low-quality input may still require review.
How is freshness measured?
We can track source change time, successful processing time, failed jobs and the indexed version. The acceptable delay is agreed for each source and use case.
Which sources should be searchable first?
Bring representative files, source locations and your update requirements.
