Data
From source to decision,
in three layers.
Each connector is defined by declarative configuration rather than code. Raw data is preserved immutably, which makes it possible to reprocess when a business rule changes instead of losing history.
Connectors
592 systems in the integration catalog.
A credentials form, the secret stored in a vault, incremental synchronization and safe reprocessing. From activation onward, data is loaded into the unified layer.
Sources in production
- HubSpot
- Jira
- InHire
- Google Drive
- Gmail / Workspace
- Google Calendar
- Google Analytics 4
- Trello
- LinkedIn Ads
- LinkedIn Pages
HubSpot, Jira, InHire and Google Calendar are connectors built in house, because their APIs require specific handling of pagination, cursor control and volume that a generic connector does not cover.
The curated catalog, by domain
- Ads & Marketing · 15
- Dev & Projects · 14
- Finance / ERP · 12
- Databases & BI · 11
- CRM & Sales · 10
- Communication · 9
- Social & Content · 8
- E-commerce · 7
- Files & Storage · 6
- Product analytics · 5
- HR / ATS · 5
- Support · 4
The ingestion engine uses permissive open source licences across the entire critical path. There is no contractual restriction on reselling the product or offering connectors to your own clients.
Normalization pipelines
Raw, normalized and consolidated.
Raw
The content exactly as received from the source, partitioned by client and by origin, in an immutable, append-only format. Re-running a synchronization does not duplicate records, and previous versions remain available.
Normalized
Canonical entities, idempotent per origin. This is the layer of identity keys — company and tax identifiers with check-digit validation, normalized email addresses, phone numbers in international format, and professional profiles. The same rules run in the transformation and in the tests.
Consolidated
Tables served by the platform, with per-client isolation enforced in the database: CRM, talent and projects. This is the layer applications and search read from, with one record per origin and reloads that do not duplicate.
Identity resolution as a domain in its own right
Approximately 70% of the unification effort consists of determining that two records represent the same entity. The platform addresses this with a dedicated data model: merge candidates, field-level provenance, manual override with history and full reversibility.
The Data Audit application lets a specialist review and label pairs, calibrating the model. The precedence rules are explicit: the system responsible for a field prevails, empty fields are completed from complementary sources, and a human decision prevails over both, with date and author recorded.
Completeness verified, not assumed
A synchronization is not considered successful when the number of records loaded is lower than the count measured at the origin itself. The discrepancy halts the process and reports the failure, rather than recording completion with partial data.
The Data Lake application shows the count, the position marker and the complete history of every run, making it possible to audit what was loaded and when.
The result is a single utilization figure, consistent across finance, hour logging and the contract — with lineage traceable to the source, the date and the connector that produced it.