Recruitment · AI workflows
Skilltage
A recruitment product that connects role definition, candidate intake, evidence review, and follow-up in one working experience.
The problem
Connect role requirements, applications, and evidence review in a usable recruitment workflow while keeping people in control of the decisions.
The delivered product
A localized public site, multi-tenant organisation workspace, candidate-facing application journeys, recruiter interactions and a separate operations surface work as one system with shared typed contracts and explicit access boundaries.
Engineering & operations
The product flow and operating layer were designed together.
Each surface shares explicit contracts and lifecycle rules. Heavy document and media work leaves the request path and runs in observable, retryable workers.
Recruitment flow
- Define
- Intake
- Evidence processing
- Human review
- Interactions
Platform flow
- Cloudflare surfaces
- Typed Lambda API
- SQS workers
- Supabase / Postgres
Operating layer
- Stripe
- SES delivery
- Support
- Metering
- Retention
- Observability
Inference that produces evidence
Confirmed job requirements gate multimodal CV and cover-letter parsing. Schema-validated output becomes requirement-level evidence, gaps and uncertainty for review, with traced stages and recoverable failure states instead of an opaque model verdict.
Elastic infrastructure without idle application servers
Cloudflare-delivered frontends, Lambda APIs, SQS workers, dead-letter queues and private Supabase storage keep request handling separate from heavy processing. Application compute can fall idle between workloads while retries and idempotent consumers protect asynchronous work.
Commercial and operational depth
Stripe subscriptions and pay-as-you-go credits, entitlement checks, idempotent usage accounting, a billing portal, SES delivery feedback, support entry points, retention jobs and production telemetry make the platform operable beyond the demo.
Video interviews with explicit boundaries
Recruiters can add a welcome and ask one structured question. Candidates test devices locally, choose which recording to upload and retain deliberate control over submission. Private media processing produces transcript-grounded notes for human review without using faces, emotions, accents, personality or video frames to rank people.
Evidence
- Product
- Recruiter workspaces and candidate journeys built together.
- Control
- AI-assisted evidence supports human review.
- Operations
- Payments, usage accounting, and support included.