EdZola
AI for NGOs

TRRAIN: real-time face-recognition attendance inside the system teams already use

A check-in used to be all it took to be marked present. Using Claude Code, EdZola built face recognition and liveness verification into TRRAIN’s Zoho Creator records, so every attendance entry is confirmed at the moment it’s made.

In production36,000+check-ins verified
Speed< 2 sfrom face to confirmed check-in
Liveness4 guardsindependent checks against photos, screens and recordings

At a glance

Partner TRRAIN (Trust for Retailers and Retail Associates of India)
Programme area Programme operations: attendance across a multi-state training-centre network
What we built Real-time face-recognition attendance with liveness verification, embedded as a widget in TRRAIN’s Zoho Creator trainee and batch records
Built with Claude Code for design, build and testing. Claude for architecture, test design and documentation.
Runs on Zoho Creator, a Python face-recognition service and AWS Rekognition Face Liveness
Status Live in production at TRRAIN centres, on personal phones and shared centre devices

The challenge: a check-in was all it took

Across training centres, attendance creates a surprising amount of invisible work: marking records, reconciling them and following up on gaps. For TRRAIN, attendance also feeds the programme reporting that funders and CSR partners rely on.

  • Nothing confirmed who checked in. A check-in was accepted as soon as it was submitted. Nothing confirmed that the person was the enrolled trainee, at any centre, on any day.
  • A proxy looked exactly like the real thing. A friend checking in on someone’s behalf left the same record as a genuine check-in.
  • Real-world conditions. Check-ins came from personal phones and shared devices, in every kind of lighting and over unreliable networks. Any verification had to work across all of it.
  • Two systems out of step. A separate attendance platform didn’t sync in real time, so changes in enrolment status didn’t reliably reach attendance, and the team spent time reconciling the two.

The solution: verification built into the check-in itself

Instead of adding another app, EdZola embedded attendance directly in TRRAIN’s Zoho Creator application. Centre users open it from the system they already work in. There’s no separate login or device.

1 · LookThe trainee faces the camera; location is captured for the session.
2 · Prove you're realA blink or a smile confirms a live person is present before any image leaves the device.
3 · RecognisedThe face is matched against the trainee's enrolled profile in under two seconds.
4 · RecordedStatus, time, GPS and the live photo are saved to Zoho Creator straight away.

What it looks like

Face attendance check-in screen on a phone showing the liveness prompt, the detect, liveness, verify and post steps, and a check-in confirmation

Check-in on a phone: liveness prompt, four-step verification and instant confirmation.

Attendance record in Zoho Creator showing date, zone, Present status, live captured image, check-in and check-out times, source, liveness action, GPS fields and distance from centre

The attendance record in Zoho Creator: status, times, live photo, liveness action, GPS and distance from the centre.

Faces, names, IDs, centres and coordinates are masked.

Four guards before anyone is marked present

On-device challengeBlink or smile detection runs in the browser.
Passive anti-spoofingEvery frame is checked for printed photos, screens and masks.
Cloud quality checkAWS Rekognition gives a second opinion in difficult lighting.
Active livenessAn AWS Face Liveness challenge for high-risk attempts. A photo or recording can't pass it.

Built for real centre conditions

  • Location-verified. GPS is captured at check-in and check-out and compared with the trainee’s assigned centre inside Zoho Creator, where the operations team can adjust the rules themselves.
  • Never loses a record. Check-ins are confirmed instantly and queued, then delivered to Zoho Creator with automatic retries, even on slow connections.
  • In step with enrolment. Daily reconciliation with TRRAIN’s trainee records means status changes take effect automatically. There’s no second system to maintain.
  • Gets better with use. High-confidence check-ins keep each trainee’s profile current as appearance, lighting and cameras change.

How Claude was used to build it

  • Claude Code for the build. The Python verification service, the Zoho Creator widget, the multi-layer liveness decision flow, the outbox queue and retry logic, and the Zoho Creator REST integration were written and tested with Claude Code.
  • Claude for design decisions. Model selection, matching thresholds, the escalation path between liveness checks, and the memory limits of a low-cost server were worked through with Claude.
  • Claude for quality and handover. Test-case design, edge cases such as HEIC and rotated enrolment photos, and the architecture and user documentation given to TRRAIN.

At runtime, face matching and liveness checks run on dedicated computer-vision models, chosen for speed and accuracy on low-cost devices.

Results: attendance TRRAIN can stand behind

The system has been live for the last few months and has verified more than 36,000 check-ins. TRRAIN now:

  • Confirms every check-in against a live face, instead of treating the check-in action alone as proof of presence.
  • Sees attendance in real time inside its programme system, without reconciling two platforms.
  • Gets automatic flags for check-ins that don’t match a trainee’s assigned centre, and for liveness failures.
  • Stops check-ins from trainees whose enrolment status has changed.
  • Keeps a verified attendance record across its multi-state network, with a photo and location trail that holds up in donor and CSR reporting.
Handled with careFaces are stored as numerical profiles that can't be turned back into a photo. Enrolment photos are processed in memory and discarded. The live photo is kept only on the attendance record in Zoho Creator. The result supports human oversight. It's a verified record, not an automatic judgement.

What other teams can learn

Verify at the moment of capture. Checking attendance later is reconciliation. Checking it at check-in makes the record reliable from the start.

Design for the device in the trainee’s hand. Layered checks keep the common case fast and cheap, and save the strongest checks for the moments that need them.

Keep the rules where the team can change them. Proximity limits and follow-ups live in Zoho Creator, so TRRAIN’s operations team adjusts them without waiting for a developer.

Start with one part of the work that feels harder than it should.

Bring one tracker, form or workflow. We’ll help you find a practical first step, whether that means a clearer process, a connected system or better use of what you already have.

Prefer email? [email protected]

EDZOLA NEWSLETTER

Hear more about the work, once a month.

Subscribe for practical ideas, field notes and updates from EdZola’s work with social-impact organisations.

A monthly email from EdZola. You can unsubscribe at any time.