How it works

From chaotic inbound to a monitored submission.

One orchestrator routes work between six independent agents, retries failures, and pauses for human approval at exactly two control points.

The six-agent pipeline

Each agent does one job and hands off through shared stores — they never call each other directly. Click any agent to see its inputs, outputs, and the systems it touches.

I
01 · Agent

Iris

Ingestion & Dedup

Collapses every vendor blast into a single canonical req. Duplicates become routes, never chased twice.

EmailTelegramDiceWhatsApppgvector dedup
In
Chaotic inbound — email, Telegram, Dice, WhatsApp
Out
One clean, scored, deduplicated job
Auto-playing the pipeline — click any agent to pause
SyncTalent.ai agent pipeline architecture: inbound email, Telegram, WhatsApp, Dice, Slack and Teams feed Iris for ingestion and dedup, then Maya for sourcing and shortlist, Theo for outreach, Aria for screening, Sam for submission and packaging, and Noor for monitoring — all coordinated by a Temporal orchestrator, with human control points at routing and escalation
The whole pipeline on one screen. The orchestrator holds state; the agents never call each other directly.

How candidates are matched

Matching is not keyword search. Deep Semantic Match embeds the résumé and the requirement and compares meaning, so a candidate whose relevance is structural rather than lexical still surfaces — and every score comes with the evidence behind it and the gaps it could not evidence. Hard filters like work authorization run before scoring, never as a weight inside it.

Deep Semantic Match mapping résumé phrases to requirement concepts: built ledger reconciliation at scale maps to payments domain, led a six-person platform pod maps to a tech-lead signal, migrated a monolith to services maps to distributed systems, and a missing Kubernetes skill is flagged as a named gap for screening
Meaning mapped to meaning, with the gaps named rather than hidden.

How candidates are screened and verified

Aria can screen by AI voice call or hand off to a human recruiter, capturing rate, availability, and work authorization into the same structured scorecard either way. For roles that warrant it, an AI interview runs with live proctoring — identity verification, second-device and tab-switch detection, assistance flags — and the scored, timestamped scorecard is attached to the submission.

AI interview with live proctoring: identity verified by government ID and face match, integrity signals showing zero tab switches, no second voice, zero paste events and two flagged off-screen glances, and a scorecard rating technical 8.4 and communication 7.9, attached to the submission
Every interview becomes a scored, timestamped record on the submission.

Where a human steps in

Automation handles the repetitive work end to end. People are pulled in at the two moments where judgment actually changes the outcome:

Onboarding in four steps

  1. Connect your ATS. Link Bullhorn (via Merge.dev) so reqs and submissions sync both ways.
  2. Import your bench. Bring in your internal consultants so Sourcing can match them first.
  3. Point your inbound. Forward vendor email and chat sources so Ingestion can dedup into canonical reqs.
  4. Go live. Watch time-to-redeploy on day one, with a human in the loop at the two checkpoints.

Want to see it on your own reqs? Schedule a demo → or explore the live demo.