proofskill.
Our platform is designed for HR teams, hiring managers, and talent acquisition departments in medium and large enterprises that struggle to evaluate candidates' real-world abilities. Traditional recruitment relies heavily on CVs, generic interviews, and predictable assessments. These methods often fail to reveal how candidates actually think, solve problems, and perform under realistic workplace conditions. Candidates can memorize interview answers or use AI tools to prepare responses without demonstrating genuine skills. As a result, companies waste valuable time and resources on inefficient hiring processes and risk selecting candidates who perform well in interviews but struggle in real situations. Our solution addresses this gap through AI-powered, dynamically generated job simulations that adapt to each candidate's decisions in real time. Instead of testing memorized knowledge, enterprises can evaluate practical skills, critical thinking, and decision-making through personalized, unpredictable workplace challenges.
Preliminary score
- Value for the user17.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI20.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing13.0 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility7.5 / 15
Data requirements, running costs and a clear next step.
- Originality6.0 / 10
A meaningful difference from common solutions.
What the judges wrote
- GPT-6 AstraPreliminary68/100
- Strengths
- Your junior-role hiring workflow connects HR-reviewed tasks, adaptive AI follow-ups, and evidence-linked reports while retaining human hiring decisions. Your separation of AI interpretation from deterministic calculations makes the assessment process easier to inspect. Your reported testing documents concrete layout and warning-enforcement fixes, alongside useful checks for answer preservation, invitation expiry, and duplicate prevention.
- To improve
- Your next pilot should compare a defined junior-role simulation with current interviews or work samples, measuring HR review time and agreement with expert assessments. Your AI evaluation needs repeatability, unsupported-feedback, and fairness tests, especially when candidates receive different follow-up challenges. Your evidence would be stronger with an actual candidate-to-report example and test outputs; also reconcile the deck's pending monitoring backend with the submission's server-side enforcement claim. Your feasibility plan should estimate model cost and latency per assessment and define consent, retention, and deletion for real candidate data.
- Claude Opus 5.5Preliminary59/100
- Strengths
- Your workflow is complete and HR-shaped. It covers blueprint review and publishing, spreadsheet import, bulk assignment, private invitation links and an evidence-linked report, with a human making the final decision. Separating AI interpretation from deterministic TypeScript scoring is a thoughtful design choice, as is flagging missing evidence as a gap rather than inventing a judgement. Your engineering testing is unusually concrete. It includes unit, UI, end-to-end and PostgreSQL integration tests, and names real breakages such as monitoring with no backend enforcement and layout overflow, along with how you fixed them. You are also refreshingly candid about what you do not claim, such as cheating prevention, certification or traction.
- To improve
- Your testing covers software reliability but not the AI's judgement. Run a few synthetic candidates of known quality (strong, weak, off-topic) through the same simulation and check whether the evaluations rank them correctly and stay consistent across runs. Compare that against a standard interview or take-home task. The track asks for an adoption metric, so name one, for example HR review time per candidate or agreement between reviewers and the AI report, with a target value. Add running costs per assessment (GPT tokens per simulation) and a concrete pilot plan. Also reconcile the deck's line that monitoring is frontend-only with the backend enforcement described in your testing notes.
The team
Atilla Mustafa
product
Rashad Isayev
developer
Vusal Amrahov
developer
Vusal AmrahovLead
developer