Introduction

Challenges

beBee has been using AI heavily to analyze data and processes formats in resumes and job descriptions submitted to the platform, with the goal of optimizing the matching of candidates and recruiters’ needs and the user subscription conversion.

Optimizing Output Performance with JSON Format

beBee needed JSON-formatted outputs for downstream tasks in their workflow. However, creating structured JSON outputs would slow down inference speed.

They had previously tried solving this by prompting the LLM, but the results were inconsistent. This became a significant block in their product development process. They were motivated to find a permanent solution.

Cutting Cost While Maintaining High Performance

Yago de la Rica, AI product manager at beBee, pointed out:

“Our traffic has been growing rapidly, which has made controlling costs a top priority for us. We have been exploring model API providers that meet our stringent requirements for both pricing and performance. To be honest, our needs have been quite challenging to satisfy.”

Rising costs due to increasing large amounts of data to be processed and the need for extensive manual post-processing to maintain content quality are significantly straining beBee’s budget. A solution that could offer both affordability and performance is critical to achieving beBee’s business goals.

Solutions

Stable and Fast Inference with 99.5% Uptime

Novita AI promptly formed a technical team to upgrade the inference engine. By integrating advanced JSON output technology and introducing multithreading support, the team fully utilized the multi-core capabilities of the CPU, significantly boosting overall throughput and reducing latency, enabling faster interactions.

Highly Performant APIs and Affordable Pricing

With consistent optmization efforts, Novita AI is able to deliver competitive pricing while ensuring top-notch output quality and performance for beBee and other customers. Our team offers one of the best prices for LLM APIs, which can be referenced on OpenRouter.

For beBee, even with surging business, our pricing stays affordable, aiding smooth scaling.

Results

3x Processing Efficiency

With Novita AI’s inference optimization, the inference speed improved by 300%, enabling beBee to more than double its daily job description handling capacity and achieving more consistent efficiency.

50% Cost Reduction

Novita AI’s model APIs not only meet beBee’s stringent performance and output quality requirements but also enable the team to reduce overall costs by 50%.

25% Increase in Paid Conversion Rates

With improved semantic analysis and keyword extraction, Novita AI significantly increased the accuracy of job-candidate matches. The improved matching precision resulted in more connection requests from job seekers and has increased the premium membership conversion rate by 25%.