Automatic Recruitment

Inspiration

Online recruitment, especially within Big Tech, is currently plagued by two major issues: rampant cheating during remote assessments and prevalent "resume padding" (exaggerating skills). When candidates overstate their capabilities to bypass automated ATS filters, honest and highly skilled applicants are unfairly pushed out of the hiring pipeline.

While remote hiring is the undeniable future of the industry, there is a glaring lack of comprehensive solutions that guarantee a fair, transparent evaluation. We were inspired to build a system that empowers HR to find genuinely talented individuals based on real capabilities, while effectively closing the loopholes that dishonest candidates exploit in online recruitment.

What it does

Automatic Recruitment is an end-to-end platform designed to completely overhaul the hiring pipeline. It automatically filters CVs and conducts dynamic technical testing to verify if a candidate's actual skills match their resume claims.

By strictly detecting and filtering out resume padding and cheating attempts, the system restores opportunities to honest, capable candidates. For employers, it ensures a much higher quality of hires, drastically reduces the resources and costs needed for large-scale recruitment, and guarantees transparency.

At the true core of our platform lies an advanced, LLM-driven evaluation benchmark and dynamic test generation engine. This system automatically crafts highly personalized technical assessments based on a candidate's real-world portfolio and tech stack. To complement this rigorous evaluation pipeline and ensure absolute integrity, the platform is further fortified by a robust AI-powered anti-cheat proctoring system.

How we built it

Our solution is a multidisciplinary architecture spanning AI, OS-level programming, a web portal, and a desktop application.

CV Ranking & Queueing: We implemented advanced LLMs to evaluate and rank CVs against specific benchmarks, creating a priority queue for the testing phase.
Personalized Assessment: For the top-tier candidates, the system generates customized technical tests. These are dynamically created based on their specific tech stack, GitHub projects (extracted via the Tinyfish API), and the company's hiring requirements. We encode crucial keywords and project pipelines to create a highly accurate evaluation criteria.
Anti-Cheat Desktop App: We built a dedicated desktop application using Electron and JavaScript. It actively monitors candidate behavior using an AI-powered camera system (incorporating face recognition, anomaly detection, and speech recognition). Furthermore, we implemented OS-level interventions to aggressively restrict unauthorized background processes and potential cheating software.

Challenges we ran into

Developing a comprehensive anti-cheat and AI system came with significant hurdles:

Resource Constraints: We faced limitations with API/testing credits.
Technical Roadblocks: Navigating strict OS-level permissions was difficult, and initially, our AI models did not perform as accurately or efficiently as desired.
Advanced Cheating Methods: We encountered hardware-level cheating techniques. For example, while we successfully handled application-level screen OCR bypasses, dealing with cheats that read directly from the GPU buffer—which cannot be entirely blocked at the OS or application level—was a massive challenge.
Evaluation Pipeline: Designing an automated question-generation pipeline that is strictly fair, technically accurate, and practically useful for HR professionals requires extensive tuning and real-world validation.

Accomplishments that we're proud of

We are incredibly proud to have built a cohesive system that successfully addresses almost all the major challenges we set out to tackle. Despite the complexity, we managed to integrate LLM-based personalized testing, computer vision proctoring, and OS-level process management into a single, functional pipeline. While deep hardware-level interventions and extensive real-world benchmarking remain ongoing, we successfully mitigated the vast majority of software-level cheating and resume-padding tactics.

What we learned

This project was a massive learning curve. We honed our skills in step-by-step problem-solving, complex system architecture, and AI model optimization to reduce server loads.

Fascinatingly, we learned a great deal about the vulnerabilities of existing anti-cheat solutions like Safe Exam Browser (SEB), which are heavily restricted by application-layer limitations. We also gained deep insights into how modern cheating agents operate to bypass human proctors—such as using screen-share bypass techniques where cheating overlays are visible to the candidate but completely hidden from the HR monitor.

What's next for Automatic Recruitment

Moving forward, we plan to enhance our AI models to be even more generative and accurate. We want to refine our evaluation pipelines and benchmarks to generate highly comprehensive, multi-dimensional assessments. A major technical goal is to develop countermeasures against even the most sophisticated, stealthy cheating methods running locally on candidate machines.

Ultimately, our vision is to establish Automatic Recruitment as the ultimate, transparent, and automated bridge connecting top-tier talent with the right employers in the growing digital economy.