Roadmap · not built yet
The scoring layer we want every ad call to pass through.
A 0-100 composite brand-safety score, built on Idukki’s own UGC moderation workflows. This is the design target, not a shipping product — there is no live inference endpoint, no SLA and no SOC 2 claim to make about it today.
Roadmap
Six-dimension score
Six-dimension score
Per-axis sub-scores
Custom profiles
Per-buyer +
Edge inference
Multi-region inference:
- 0-100Composite score per call · the target shape
- 6 axesToxicity · hate · NSFW · IP · brand · deception · target axes
- 0Customers on it today
- TBDTimeline · tell us if you need this next
- 01
Six-dimension score
Toxicity, hate, NSFW, IP-infringement, brand-conflict, deception — the axes we’re designing for. Per-axis or one composite, your call, once built.
- Per-axis sub-scores in 0-100: design target
- Threshold-based hard/soft fail: design target
- Why-it-scored explanations: design target
- 02
Custom profiles
The plan: a per-buyer profile, which categories to block, warn or allow, pushed from a dashboard or the API. Not built.
- Per-buyer + per-campaign profiles: planned
- Dashboard + API: planned
- Profile versioning + audit log: planned
- 03
Edge inference
The intent is low-latency inference close to the request. No deployment exists yet — no Cloudflare Workers, Fastly or Lambda@Edge integration has been built.
- Multi-region inference: not started
- CDN-cache friendly response shape: not started
- Uptime SLA: none to offer yet
- 04
Pre-bid hook
The idea: score before the bid request leaves the SSP, so blocked impressions cost nothing. No OpenRTB integration exists.
- OpenRTB ext.brand_safety field: proposed
- Pre-bid latency target: not measured, nothing runs yet
- SSP compatibility: not started
- 05
Post-serve audit
The plan: log every served impression, re-score retroactively on creative change, roll up to reports. None of this exists yet.
- Per-impression score storage: not built
- Retroactive re-scoring: not built
- Automated reports: not built
- 06
Custom training
Longer-term idea: bring your own labelled examples and we fine-tune a private head. No base model exists to fine-tune on top of yet.
- Bring your own labels: future idea
- Fine-tuning pipeline: does not exist
- Private inference endpoint: does not exist
- Research
- Building
- Beta
- Live
Tell us if this would unblock a deal
This pillar is the least-built part of the API platform roadmap. If brand-safety scoring is the thing standing between you and using Idukki, say so — that’s what decides build order, not a beta that doesn’t exist yet.
- No free trial or API key to offer today
- Real conversation with the platform team, not a form-fill
- Tell us your volume + use case and we’ll tell you honestly where this sits on the roadmap
We read every message and reply by email. If it fits, we suggest a call.