Client
Resume screening platform, United States
Engagement
Infrastructure migration and matching engine rebuild
Team
Jellie engineers, sole engineering team
The starting point
The platform did its job: recruiters uploaded resumes, the system evaluated them against open roles. But the whole evaluation layer ran on a single prompt, and the infrastructure underneath had been assembled quickly in the product’s early days.
That setup carried three problems. Results were inconsistent, because one prompt was doing the work of an entire scoring system. Costs scaled linearly with volume, with no way to optimize. And every product improvement meant editing a prompt everyone was afraid to touch.
What we built
Infrastructure first
We migrated the platform to AWS with proper separation between ingestion, processing, and serving. The migration ran alongside the live product; users never saw downtime.
The matching engine
We replaced the single prompt with a scoring platform: structured extraction from resumes, evaluation broken into discrete scored dimensions, and explicit logic for how dimensions combine into a final ranking. Scores now come with reasoning a recruiter can read and challenge.
Built to be tuned
Scoring weights and criteria are configurable per role type instead of buried in prompt wording. The client’s team can adjust how the system evaluates without touching code.
The outcome
A screening platform that produces consistent, explainable scores on infrastructure that scales with volume. The client's team owns the system fully, tunes scoring without touching code, and ships improvements without us. That last part is the point.