A Technical Walkthrough of What Happens When a New Hire Clicks the Link

Aug 20, 2026 | Blog

Reading Time: 4 minutes

A new joiner gets an email. Inside it, one link. They click it on their phone while still holding their morning coffee, half expecting a login screen, a password reset, the usual friction of joining anything corporate. Instead, in well under a minute, they're looking at Priya's face, and she already knows their name.

Nothing about that moment feels technical. Almost all of it is. Here is exactly what happens in the gap between the click and the hello, step by step, system by system.

0sThe link landsOne link in an email, opened on a phone0 to 2sToken checked, not trustedValidated against the database for expiry and one-time use2 to 5sPage loads with data attachedThe joiner is already known before anything renders5 to 15sThree services wake upACS, ElevenLabs and D-ID start in parallel15 to 30sThe voice gets its briefingRole, start date and personalised script loaded30 to 45sAudio and face syncLip-synced avatar streams over WebRTC47sHelloShe already knows their name
Under a minute, and most of it is three unrelated services agreeing to behave as one.

Second 0: The Link Lands

The URL contains a single UUID token, generated by HR back when the invitation was created, tied to one candidate record and nothing else. There's no username, no password, nothing to type. The joiner's browser sends that token straight to the API.

Seconds 0 to 2: The Token Gets Checked, Not Trusted

The token hits a dedicated, publicly accessible endpoint built specifically for this one purpose, and nothing else on the system trusts it blindly. The database checks two things: has this link expired, and has it already been used. Either failure stops everything right here, no call, no exceptions. If it passes, the system flips the one-time-use flag immediately, this exact link now works exactly once, ever, and issues a short-lived JWT scoped only to this candidate and this call.

Seconds 2 to 5: The Page Loads with Data Already Attached

The call page itself renders server-side with the candidate's context already embedded, their name, role, department, and which call segments HR selected for them (maybe a portal walkthrough, maybe an income tax demo, always an intro and a closing). This gets handed to the page's JavaScript through a small embedded JSON block rather than a separate API round-trip, shaving a network request off the critical path before the call has even started.

Seconds 5 to 15: Three Services Wake Up at Once

This is where it gets genuinely parallel. The browser opens a SignalR connection to the OnboardingHub for real-time events. Simultaneously, it requests an Azure Communication Services token, scoped and short-lived, to join the actual audio and video session. If HR enabled the avatar for this particular joiner (a per-candidate toggle set at invitation time), a third connection begins negotiating a WebRTC peer connection with D-ID for the avatar video stream. None of these three wait for each other, they establish concurrently, because making a nervous new hire watch three separate loading spinners in sequence would undo half the point of the experience.

Seconds 15 to 30: The Voice Gets Its Briefing

Before Priya says a single word, the system opens a session with ElevenLabs' conversational AI agent and injects the candidate's full context directly into its system prompt. This is the step that turns "hello, welcome" into "hello Priya, welcome to MITS, I understand you're joining as a Full Stack Engineer on the fifth" (names changed, but the personalization is real, extracted from the candidate's uploaded offer document weeks earlier by GPT-4o).

Seconds 30 to 45: Audio and Face Sync Up

ElevenLabs begins generating speech. D-ID, listening to that same audio stream, begins mapping it to facial animation in real time, mouth shapes, expressions, timed to the millisecond. The two services were never designed to talk to each other, they're synchronized entirely through the orchestration layer MITS built to sit between them.

Second 47: Hello

The avatar's mouth moves. The words land in sync. HR, sitting quietly on the same call, sees the joiner's profile card and segment checklist appear on their dashboard at the same instant. Everyone is now present, together, and the new joiner has no idea that four independent cloud services, a token validation, a database write, and a personalized AI briefing all happened in the time it took them to sit down.

That's the whole trick of good engineering. It's not that nothing complicated happened. It's that none of it was allowed to show.

---

MITS Global Consulting builds systems where the complexity lives in the engineering and never in the experience. Explore more at https://www.mitsit.in/

MITS AI Onboarding · part 6 of 6Part of a 6-part series on the MITS AI Onboarding project.