TL;DR - Emissary enabled Rockstar to build a classification language model for candidate profile review to replace their existing GPT5 + Deepseek pipeline. The model reduces false negatives by 15%, reduces cost by 80+%, while operating at 100X the speed, allowing clients to review better results in under 24 hours instead of 3-5 days and reducing CoGS impact of review from 30% → <5%.

Introduction

Rockstar is an AI-native recruiting platform that automates talent screening and placement at scale. The platform blends humans with AI to deliver world-class recruiting support for less than $1.5K per role, with fully customizable screening scripts that listen to candidate answers to determine fit. Rockstar helps companies across industries rapidly identify and hire top talent, serving millions of candidate evaluations monthly.

Challenge

A critical bottleneck in Rockstar's AI workflow was reviewing candidate profiles against hiring criteria and determining advancement - a decision made millions of times daily that directly impacts both placement accuracy and speed-to-hire. Traditional generative LLMs proved inadequate for this specialized task, forcing Rockstar to contend with persistent friction points:

• Inconsistent classification accuracy despite prompt engineering and model switching across frontier closed and open source models.
• Rate limiting constraints that extended pipeline review cycles, forcing clients to wait 3-5 days to see full candidate pipelines and creating friction in the hiring process.
• Escalating inference costs consuming 30% of cost of goods sold through both token expenses and substantial server infrastructure needed to handle computational demand as they invested in larger models and longer prompts to handle edge cases

Solution - Emissary Classification LM

Investment: ~4000 samples ; 2 weeks.**

Rockstar post-trained a Classification Language Model tailored for profile review verification. Leveraging their proprietary candidate dataset, Rockstar's founder himself fine-tuned a 8B base model on Emissary's platform using just ~4000 samples over two weeks, leveraging Emissary's custom classification LM training technique optimized for high-precision decision-making.

Emissary also handled the full ML infrastructure across training and inference - GPU selection & provisioning, access to latest base models, templated training workflows, and model hosting - enabling Rockstar to concentrate on what matters most: their proprietary evaluation criteria and domain expertise. The model was deployed via Emissary's serverless inference for immediate, cost-effective integration, then scaled to dedicated scalable inference as traffic ramped into production.

Throughout the engagement, Emissary's team provided guidance on configuration choices, confidence calibration strategies, and deployment optimization, ensuring reliable performance while maintaining Rockstar's aggressive timeline.

Impact

Rockstar's proprietary classification model delivered transformational results:

• Superior Accuracy: 15% reduction in false negatives when benchmarked against an ensemble pipeline comprising GPT5 + Deepseek.
• Dramatic Time to Value Improvement: Latency reduced by 90% to 0.15s/classification, resulting in time-to-first-value reduced from 5 days to 24 hours.
• 80% Cost Reduction: 1/5th the inference cost of their previous approach, reducing impact of this workflow on CoGS from 30% to under 5%.
• Confidence Scoring & Calibration: The model now provides continuous confidence scores with improving calibration as data evolves, enabling built-in candidate ranking and smarter decision making over time.
• Elimination of Rate Limits: No throttling - already supporting 1M+ requests monthly with autoscaling infrastructure and rapid growth with no upper bound, ever.

What’s next?

Building on this success, Rockstar is expanding deployment of Emissary-powered models across their platform to include developing customer and role-specific decisioning models, allowing hyper-personalized candidate screening that adapts to each company's unique hiring criteria and culture gleaned from hiring manager feedback, cementing Rockstar's position as the most technically sophisticated AI-native recruiting platform in the market.