Growlr: A Dating App Built on Verified, Safe Content

Client: Growlr

Industry: Dating Apps 

Capabilities Demonstrated: Image Moderation, Age & Identity Verification, Fraud Detection, Community Guideline Enforcement 

Deployment: Full production integration in 4 weeks

The Environment

Growlr runs a high-volume, image-first community where trust depends on every profile being a real, consenting adult who belongs on the platform. That creates a narrow but demanding moderation brief: every uploaded image has to be screened before it’s visible to anyone else, underage content has to be caught and escalated with zero tolerance and profiles that violate community guidelines fake accounts, scam images or content outside the platform’s stated audience need to be identified and actioned fast.

Prior to deploying Foiwe’s engine, Growlr’s internal telemetry showed a familiar pattern for image-heavy platforms at scale: a meaningful share of uploads required manual review, response time on flagged content lagged behind upload volume, and fraud/spam accounts were often caught only after user reports accumulated.

The Engine: What Foiwe Run

Foiwe deployed an image-first moderation and fraud pipeline integrated directly into Growlr’s upload flow, so every image is scored before it publishes.

Image Moderation & Classification

• Explicit/nude image detection across profile photos and shared media, with automatic hold-and-review before publication
• Zero-tolerance CSAM (child sexual abuse material) detection any suspected match is instantly blocked, the account is suspended, and the case is escalated through mandatory reporting channels (e.g., NCMEC) per legal requirements. This layer runs ahead of every other classifier in the pipeline.
• Age-signal classification to flag any content suggesting a user may be under 18, routing those accounts for immediate suspension pending verification
• General image classification (scene, content category, policy category) to route borderline uploads to the correct review queue

Fraud Detection

• Behavioral and image-pattern signals to catch scam accounts (stock-photo reuse, romance-scam patterns, payment-redirect language paired with mismatched imagery)
• Duplicate/reused-photo detection across the profile base to catch bot and catfish accounts

Community Guideline Enforcement

• Automated detection and removal of profiles and images that fall outside Growlr’s stated community and audience guidelines (e.g., female-presenting profiles/images on a platform built for men), routed for policy-based takedown rather than manual audit
• Repeat-offender tracking to prevent banned content or accounts from re-entering under new uploads

The entire pipeline runs pre-publication: nothing goes live until it clears the classifier stack.

Deployment Speed

Phase	Timeframe	Focus
Onboarding & Integration	Week 1–2	Engine integrated into Growlr’s image upload pipeline
Calibration	Week 3	Threshold tuning for explicit content, age-signal classification, and guideline-enforcement rules
Full Rollout	Week 4	Full classifier stack live across all new and existing uploads
Optimization	Ongoing	Continuous model tuning from flagged-content feedback loops

Engine Output: Volume, Accuracy, Speed

Metric	Pre-deployment Baseline	Foiwe Engine Output
Image review turnaround	Manual/hybrid, hours	Automated, under 5 seconds per image
Underage-signal / CSAM escalation	Manual/report-driven	Automated pre-publication block + mandatory escalation
Fraud & duplicate-photo accounts	Report-driven detection	Detected pre-publication via image + behavior signals
Guideline-violating profiles (off-audience content)	Manual audit	Automated detection and removal
Explicit content reaching users	Post-report takedown	Blocked at upload, before publication

Read on the engine, not the narrative:

• Zero-tolerance layer runs first — CSAM and underage-signal detection execute ahead of every other classifier, so no other rule can allow that content through; matches trigger immediate account suspension and mandatory reporting.
• Pre-publication moderation — because scoring happens before an image ever publishes, exposure to other users is prevented rather than remediated after the fact.
• Throughput — sub-5-second classification per image at full upload volume, not a sampled batch.
• Guideline enforcement is automated, not audited — profiles and images outside Growlr’s stated platform audience are identified and actioned by the same pipeline, rather than requiring a separate manual sweep.

Why the Approach Holds Up

1. Safety-first pipeline ordering — the CSAM/age-signal classifier sits ahead of every other rule in the stack, so this category is never subject to the same thresholds or tuning applied to general content classification.
2. Pre-publication scoring — moderation happens before content is visible to any user, closing the exposure window that report-driven systems leave open.
3. Unified image + behavior signal — fraud detection combines image classification (duplicate/stock photos) with behavioral patterns, catching scam accounts that either signal alone would miss.

Summary

Deployed on Growlr’s upload pipeline, Foiwe’s engine moved image moderation from a manual, report-driven process to automated pre-publication screening with a zero-tolerance classifier for CSAM and underage content running ahead of every other rule, sub-5-second turnaround on general image classification and automated enforcement of platform community and audience guidelines.