01
Face Detection
Find the face in each capture.
Locates the face and its landmarks, and refuses a frame with no face or with more than one.
Inside every verificationFace identity & verification platform
Computer vision infrastructure,
built for the real world.
A secure API for 1:1 face verification with presentation-attack screening, now in restricted pilot — built with the environments our region actually uses in mind.
API-firstPrivacy by designEngineered in Iraq
Illustrative face-mesh visualization — not a biometric capture
InterfaceOne versioned API
FormatREST · JSON
DeploymentHosted pilot
StatusRestricted pilot
The platform
One API. Multiple computer vision capabilities.
One authenticated API. Every check below runs inside enrollment or verification, so there is a single integration to build.
02
Face Verification
Compare a fresh capture with an enrolled identity.
One-to-one matching against an enrolled identity, with a similarity score and a provisional threshold. During the pilot a person reviews every match.
POST /v1/identities/{id}/verify03
Liveness Detection
Screen for presentation attacks.
Passive screening on every verification, with multi-frame checks for replay. Not yet validated on attack media, so it can't approve anyone on its own.
Inside every verification04
Face Quality
Evaluate whether a capture is good enough to trust.
Blur, lighting, contrast, pose, and face size are checked, so bad captures are refused before they become bad decisions. Occlusion isn't checked yet.
Inside every verification05
Identity Enrollment
Securely enroll a biometric identity.
Quality-gated enrollment that stores an encrypted face representation — not a photo album — isolated per tenant.
POST /v1/identities06
Face Embeddings
Encrypted templates for 1:1 matching.
Each face becomes an encrypted template used only for 1:1 matching inside Warka. Embeddings are never returned through the API.
Internal, never returned07
Anti-Spoofing
Protect verification workflows from presentation attacks.
Screening for common attacks such as printed photos and screen replays, surfaced as an explicit signal. It does not stop every attack.
Inside every verification08
Vision Intelligence
Future computer vision modules.
Document capture, scene understanding, and regional OCR are on the research roadmap — same API surface, same security model.
On the roadmap
One platform. An ecosystem of applications.
Face Identity Platform
/v1
Attendance systems
HR platforms
Education
Access control
FinTech
Customer verification
Your application
Employee attendance is the first workflow we're building for, but the platform itself is the product. Other applications can integrate the same API.
Interactive demo
Watch a verification happen.
A simulated walkthrough of the production pipeline. Every number below is illustrative — no camera is accessed and no real inference runs in this preview.
Face input
—
Simulated input — no camera is accessed
- 1
Face detection
Locate the face and landmark points
- 2
Quality check
Blur · lighting · pose · face size
- 3
Liveness
Presentation-attack signals
- 4
Face embedding
Compact identity representation
- 5
Identity verification
Match against the enrolled identity
Developer experience
Three calls between your app and a decision.
A single authenticated REST surface. Predictable JSON. Errors that tell you what to fix. The HTTP API is the contract, and there are no packaged SDKs.
Request
verify.sh
# 1. Your backend starts a capture session
curl --request POST https://api.warka-iq.com/v1/capture-sessions \
--header 'Authorization: Bearer YOUR_API_KEY' \
--header 'Content-Type: application/json' \
--data '{"identity_id": "idn_..."}'
# 2. The app uploads a frame with the capture token
curl --request POST https://api.warka-iq.com/v1/capture-sessions/cap_.../frames \
--header 'Authorization: Capture CAPTURE_TOKEN' \
--form 'frame=@face.jpg'
# 3. Your backend asks for the decision
curl --request POST https://api.warka-iq.com/v1/identities/idn_.../verify \
--header 'Authorization: Bearer YOUR_API_KEY' \
--form 'capture_session_id=cap_...'Response200 OK
response.json
{
"id": "ver_01m...",
"decision": "manual_review",
"reason": "restricted_mode",
"retry_allowed": false,
"restricted_mode": true,
"identity": {
"match": true,
"similarity": 0.71,
"threshold": 0.363
},
"pad": {
"status": "passed",
"score": 0.93
},
"timings_ms": { "total_ms": 180.4 }
}Console
A console built for operating identity.
Usage, keys, webhooks, and audit — the operational surface your security team will ask about on day one.
app.warka-iq.com/console
Product preview
API requests · 7 days
48,210
Successful verifications
31,876
Failed verifications
1,204
Liveness checks
33,080
p95 latency · 7 days41 ms
Verification volumeSample data
API keys
- productionwv_live_ ····Active
- stagingwv_test_ ····Active
Webhooks
- verification.completedEnabled
- identity.deletedEnabled
Use cases
Infrastructure, not a single app.
Employee attendance comes first, and the same API can verify students, customers, and visitors. During the restricted pilot, a person reviews every match in all of these.
Employee verification
Problem
Attendance and workforce systems trust cards, PINs, and GPS — all of which can be shared, forgotten, or spoofed.
Solution
Compare the employee's face at check-in, with attack screening, on managed devices or their own phones.
Integration
Your attendance backend calls identity verification, acts on the decision, and stores the audit record.
Outcome
Stronger evidence of who checked in, without new hardware.
Regional engineering
Built for the conditions we actually live in.
Vision models are trained somewhere, for somewhere. We plan to evaluate ours against this region's real environments — harsh daylight, dim offices, mid-range Android cameras, and realistic appearance variation — and that evaluation hasn't been done yet.
Regional evaluation program — plannedAndroid · DaylightIllustrative
Lighting
Camera
Pose
Pipeline response
Illustrative pipeline behavior- Strong daylight — within normal operating range
- Typical mid-range mobile capture
- Frontal pose — optimal for matching
Decision
Proceed to verification
Condition-level accuracy figures will be published when our regional evaluation completes. We don't quote numbers we haven't measured.
Appearance variation is an engineering requirement
Evaluation focusRobust face verification should hold up across realistic appearance variation — different head coverings including hijab, glasses, beards, changing lighting, camera quality, pose, and partial occlusion. We haven't measured this yet. It is what our evaluation will test, and we will publish how the system performs across these conditions rather than asking anyone to assume it.
Research
Built on research. Engineered for production.
Face verification is a measurement system. We treat every stage of the pipeline as something to be tested, calibrated, and improved — never a black box.
The verification pipeline
- Image
- Detection
- Alignment
- Quality
- Liveness
- Embedding
- Similarity
- Decision
Deep learning
Convolutional and transformer models trained for face analysis tasks.
Metric learning
Objectives that pull same-identity faces together and push different identities apart.
Face embeddings
Compact vectors that represent identity, not appearance details.
Similarity search
Fast comparison between embeddings at verification time.
Presentation attack detection
Models that try to tell live faces from photos and screens. Ours isn't validated yet.
Image quality assessment
Predicting whether a capture will produce a reliable decision.
Model optimization
Quantization and distillation so inference stays fast on practical hardware.
Threshold calibration
Choosing operating points from measured error trade-offs. Until we've measured real data, Warka uses the model's published default.
An engineering and research journey in computer vision
Warka Vision began as applied computer-vision research: building a full verification pipeline, benchmarking design choices, and studying how face verification behaves in the environments around us. The commercial platform and the research program advance together — what we learn is what we ship.
- Full-pipeline experimentation — detection to decision
- Evaluation protocol design for regional, real-world conditions
- Reproducible benchmarking before marketing claims
- Responsible-biometrics engineering practices
Measurement
Trust is measured.
A verification platform asks you to trust its error rates. We will publish ours — measured, reproducible, and updated per release — rather than asking you to trust adjectives.
FAR
—
False acceptance rate
How often a different person is wrongly accepted.
Benchmarking in progress
FRR
—
False rejection rate
How often the right person is wrongly rejected.
Benchmarking in progress
EER
—
Equal error rate
The operating point where FAR and FRR meet.
Benchmarking in progress
P50
—
Median processing time
Time Warka spends on one verification, not counting the network.
Benchmarking in progress
TPS
—
Throughput
Verifications sustained per second per node.
Benchmarking in progress
Error trade-off across thresholds
IllustrativeFalse acceptsFalse rejects
Decision threshold →Illustrative curve showing how false accepts and false rejects trade off against the decision threshold — not measured data.
What we will publish
- A public evaluation protocol anyone can inspect
- Regional test conditions, documented
- Results updated with every model release
benchmark_report.pdf·In development
Security
Biometric data deserves more than a password.
Security designed around measurable risk — controls you can verify in the product, not slogans on a landing page.
The path of a request
- Your application
- TLS
- API gateway · auth · rate limits
- Isolated tenant services
- Encrypted storage
Encryption in transit
TLS on every API call — no plaintext path into the platform.
Encryption at rest
Biometric representations and records encrypted in storage.
Tenant isolation
Each customer's identities and keys are logically isolated.
API authentication
Scoped keys per environment with least-privilege defaults.
Key rotation
Rotate credentials without downtime when policy demands it.
Rate limiting
Per-organization limits that blunt abuse and brute-force probing.
Audit logs
Every verification and administrative action leaves a record.
Secure storage
Representations stored instead of raw galleries wherever the workflow allows.
Data minimization
We store what verification needs — and no more.
Retention controls
Configurable retention windows per class of data.
Deletion workflows
Identity deletion designed to propagate through storage.
Monitoring
Health checks on every service. Alerting is being added.
No system is unbreakable — anyone promising that is selling something. Our commitment is engineering against measurable risk, transparent controls, and fast, honest communication when something goes wrong.
Privacy
Privacy by design, not by settings page.
Biometric identity is sensitive. The platform is designed around minimizing what is collected, stored, and retained.
- 1
Capture
Your application captures an image with user awareness
- 2
Validate
Quality checks refuse unusable captures
- 3
Represent
The image becomes a compact representation for matching
- 4
Store securely
Encrypted, tenant-isolated storage
- 5
Verify
Comparisons run against the representation
- 6
Retain / delete
Controlled retention and deletion workflows
Minimization
Collect and keep only what verification requires.
Purpose limitation
Biometric data serves verification — not advertising, not profiling.
Control
Retention windows and deletion are customer-controlled.
Transparency
What is stored, and why, is documented — not discovered.
Customer organizations remain responsible for lawful collection, notice and consent, and appropriate use of biometric data under the laws and regulations that apply to them. This page describes engineering design — it is not legal advice.
Why Warka
What we care about.
Built specifically for computer vision
Not a generic platform with a face API bolted on — vision is the whole company.
API-first
The product is the API. Everything else is built on top of it.
Regional deployment awareness
Built with MENA environments, devices, and constraints in mind. Formal evaluation is still to come.
Developer friendly
Predictable REST, honest errors, documentation written to be read.
Security focused
Biometric-grade controls from the first architecture diagram.
Measurable performance
We don't quote a figure we haven't measured.
Research-driven
The pipeline evolves with our applied research program.
Honest about limits
Restricted pilot mode stays on until attack detection has been validated.
Customer stories — coming with early access
We would rather show you a real deployment than a rented testimonial.
Origin
Built here. Ready for everywhere.
Engineered in Iraq.
Designed for real-world environments.
Built to serve applications anywhere.
Five thousand years ago in Uruk — modern Warka, southern Iraq — an artist carved what is among the earliest known sculptures of a human face. Teaching machines to understand faces is, for us, the continuation of a very old story. Warka Vision is engineered in Baghdad, designed with the region's real conditions in mind, and built to serve applications far beyond it.
The region we design for — illustrative map
Connected to the wider MENA regionResources
Learn how the pieces work.
Writing on computer vision, identity, and engineering for real-world conditions — publishing as the platform approaches early access.
Engineering
How face verification actually works
From pixels to a decision — the full pipeline, explained.
Coming soon
Fundamentals
Face recognition vs face verification
1:1 and 1:N are different problems with different risks.
Coming soon
Security
What is liveness detection?
Presentation attacks, and what passive liveness can and can't do.
Coming soon
Product
Why GPS alone is not enough for identity
Location proves a phone was somewhere. Identity proves a person was.
Coming soon
Research
Building computer vision for real-world conditions
Lighting, cameras, occlusion — engineering beyond the benchmark set.
Coming soon
Research
How we evaluate face verification
Our metrics, our protocol, and why reproducibility matters.
Coming soon
Give your application vision.
Start building with computer vision infrastructure designed for real-world identity verification.