Eduardo Cepeda
← All work

Aindez / AI & apps

Connecting recruitment to the systems around it.

Connectors that bring HR platforms, document extraction, identity verification, and business data into Aindez’s recruitment workspace—with explicit checks and recoverable execution.

Try the interactive explanation ↓
My role
Fullstack / AI Engineer
When
October 2025 — May 2026
Built with
Django · FastAPI · TanStack Start · Agno · React · Zod · Tailwind CSS

Explore how it works

Try it yourself ↓
INTERACTIVE EXPLANATION · SAMPLE DATA

Different systems.
One connected workspace.

Explore the connectors I built.
Choose one to see what its workflow makes possible.

AindezSample North workspace
Verify before use
TalanaPeople & HR
PEOPLE & HR

Candidate handoff

Connection required

Review first. Link the person. Deliver the approved documents with their own receipts.

01
Sample candidate 01One vacancy · North connection
Preflight
  1. 1
    Review before sendingNo delivery until both checks are complete
  2. 2
    Link the personWaiting for preflight
  3. 3
    Deliver the documentsNo documents sent
Review the sample identity and required documents before sending.

Based on authored, merged connector work. All configurations, records, events, and results here are invented local examples. No provider request, identity check, model, or file import runs from this page.

The challenge

Each external system has a different contract: credentials, document states, identity events, data models, and failure modes. I worked across the interface, API middleware, and organization service to make those differences visible and keep one connection’s state separate from another’s.

How I built it

Verify a connection before using it

I implemented provider-specific credential checks, safe error handling, and setup interfaces that require a successful check before saving. Changing connection settings invalidates that result, and instance-scoped controls keep data and operations tied to the selected integration.

Respect each provider’s workflow

I built document extraction and review flows for DocuTray, configurable identity workflows for Didit, and latest-state handling for MetaMap. API health, incoming webhook events, and human review remain separate signals. Odoo imports expose read-only business models and canonical field mappings.

Make the HR handoff recoverable

For Talana, I built human identity review and required-document checks before delivery. Person and document results are tracked separately; retry controls target the selected candidate, connection, and execution so completed work is not confused with a different handoff.

Keep imports honest

I added Metabase endpoint configuration that tests an inactive draft before activation, alongside streaming imports that block incomplete provider exports. The interface distinguishes a verified connection, a tested endpoint, and data that is ready to use.

What came out of it

Merged connector work across six providers and three application layers: setup and verification, provider-specific operations, human review, scoped results, and recovery. These contributions sit alongside my earlier recruitment, agent orchestration, and compensation-rule work.

Role dates follow my résumé; the case also includes later verified contributions. The connector workbench uses fictional connections and records. It does not contact providers or imply that a merged change has been deployed.

Sources & project context
Next projectWinemaker Travel ↗