[ 00 ] Animal re-identification / Research preview

Every animal, a digital identity.

Kyshik recognises individual dogs from a photo of their face, so vaccination teams know which dogs they have already met. The model suggests. A person decides.

Read the method Join the field beta
Species
dogs, first
First pilot
rabies vaccination campaign · Nov 2026
Decision
human, every time
Scan E-1123 · Face · 1600 pxQuality OK
Example dog photo with the detected face framed

Closest dogs in this campaign · Model v0.3

  • KY-00420.91
  • KY-00630.64
  • KY-00170.41
Is this Simba?Waiting for the volunteer
Fig. 01 — Example scan. A score is a similarity between photos, not a certainty.

[ 01 ] Method

From one photo
to a known dog.

Four steps on a volunteer’s phone, with the dog in hand. Each step leaves a record you can check later.

  1. 01 / Capture

    Photograph the dog

    Face first, then the left and right sides, and a nose close-up if the dog holds still. The scan is saved on the phone before anything else happens.

    → works offline

  2. 02 / Embed

    Turn a face into numbers

    The engine converts the face photo into a vector, a numeric fingerprint of that dog’s features, and checks whether the photo is sharp enough to use.

    → photo → embedding + quality

  3. 03 / Compare

    Find the closest dogs

    The vector is compared with reference photos of dogs already in the campaign. The three closest come back with a score from 0.00 to 1.00.

    → top 3 · two decimals

  4. 04 / Decide

    A person confirms

    The volunteer confirms the match, registers a new dog, or sends the scan to Review, where a coordinator compares the photos later.

    → the model never confirms

Example face photo of a golden retriever
Face required ✓Left ✓Right ✓Nose optional

saved on phone
10:42 · GPS ±6 m

Fig. 02 — The four photo views of one scan. Only the face is required.

Score bands · Default thresholds

< 0.60

Register a new dog

Nothing in the campaign looks close.

0.60 – 0.84

Pick from up to three

The volunteer compares photos side by side.

≥ 0.85

“Is this Simba?”

Yes, no, or not sure. Never automatic.

Fig. 03 — Thresholds belong to a model version and are set on the server, never inside the app.

[ 02 ] Principles

Rules that never bend.

The worst mistake Kyshik could make is an unvaccinated dog counted as vaccinated. Everything below exists to prevent it.

  1. P-01

    The AI never decides.

    It suggests the closest dogs with scores. A person confirms every identity and answers every vaccination question.

  2. P-02

    When unsure, send to Review.

    Identity can wait. Vaccination follows the campaign protocol, not the match.

  3. P-03

    Nothing is deleted.

    A fix is a new log entry. Every identity change records who made it, when and why.

  4. P-04

    No scan is ever lost.

    Each scan is saved on the phone first. No signal, a crash or a closed app cannot lose it.

  5. P-05

    Numbers are calculated.

    No one types a total. Every count is computed from the records, the same way every time.

  6. P-06

    Private by default.

    No owner data is collected. Public pages never show where an animal was found.

[ 03 ] The record

One dog.
Many scans.

Every dog gets a Kyshik ID and a passport. Each scan, one dog held at one time and place, adds a dated line to its history.

Kyshik ID
KY-0042, one per real dog
Scan
E-1123 · time, GPS, condition
Suggestion
saved before any check, never edited
Decision
who, role, reason, time
Vaccinated today?
Yes / No, never preselected
Ground truth
microchip, collar tag, paint mark
Simba, example dog

Passport · KY-0042

Simba

MaleSterilizedAdultalso KY-0091
Vaccinated 4 Nov 2026 · valid until Nov 2027Rabies
  • 24 Nov 2026E-1123confirmed by volunteer · no need today
  • 12 Nov 2026E-1071moved from KY-0091 by coordinator
  • 4 Nov 2026E-1009vaccinated · chip matches
Fig. 04 — Example passport. A wrong match is fixed by moving one scan; the old ID becomes an alias.

[ 04 ] Measures

Numbers that say only what the
data can prove.

coverage=vaccinateddogs scanned

Always labelled “of dogs scanned”. Real coverage may be lower, because it only counts the dogs volunteers scanned.

Def. 01

Dogs scanned

Registered dogs with at least one decided scan in the campaign. A dog and its alias count once.

Def. 02

Vaccinated

Dogs scanned with a valid vaccination, including a vaccination card or collar seen, each counted once.

Not measured

Population size

Kyshik does not estimate how many dogs live in an area, and draws no target lines.

CellScannedVaccinatedCoverage of dogs scanned
Cell A3025
83%
Cell B2419
79%
Cell C2214
64%
Total7658
76%
Fig. 05 — Example campaign for illustration, not pilot results. 18 dogs scanned were not vaccinated.

[ 05 ] Field notes

Notes on humane field work.

Principles and future plans. They do not claim every idea is already built.

All notes →

[ 06 ] Crew

The people behind Kyshik.

Building tools for humane animal records, from Kazakhstan. Күшік means “puppy” in Kazakh.

Kyshik founders planning the pilot
Fig. 06 — Planning the pilot.

[ 07 ] Field beta

Help test Kyshik in the field.

We need people to test photo quality, saved scans and offline use in their area. You never need to approach or handle an animal.

Create an account in the field app, then submit your organization and campaign details. Pilot access is enabled after review.

No sample records are added to your workspace.

Create an account

Coordinators can also open the dashboard. Use the same account in both apps.

Contact

Pilots, partnerships, volunteering, data.

Running a vaccination or sterilisation campaign? We’d like to hear what your teams need.

Contact page ↗
Based in
Kazakhstan
Languages
English · Русский