TalentProof
Target Users * SMEs and growing companies * HR departments and recruitment agencies * Employers with frequent hiring needs The Problem * Manual CV screening is time-consuming and costly. * Identifying the best-fit candidates is difficult. * Relevant skills and competency gaps are often overlooked. * Hiring decisions can be inconsistent and subjective. Our Solution TalentProof analyzes CVs against job requirements, matches relevant skills, highlights competency gaps, and supports evidence-based candidate evaluation.
Preliminary score
- Value for the user15.0 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI18.5 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing10.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility9.0 / 15
Data requirements, running costs and a clear next step.
- Originality5.0 / 10
A meaningful difference from common solutions.
What the judges wrote
- GPT-6 AstraPreliminary59/100
- Strengths
- Your project addresses a recognizable HR workflow, and the dashboard shows configurable job requirements alongside ranked candidates. Your emphasis on local inference, blind screening and evidence-linked skill matches addresses practical privacy and auditability needs. Your reported 100-sample confusion matrix supports the listed metrics, while your submission acknowledges parsing, wording and data-balance limitations.
- To improve
- Your demonstration should trace a CV upload through extraction, evidence-linked matching and ranking, and reconcile the Logistic Regression versus Random Forest descriptions. Your benchmark needs label provenance, train/test separation, concrete failure examples and a measured manual or keyword-screening baseline, including recruiter time saved. Your qualification rules need clarification: the screenshot marks a six-month-experience candidate qualified for a role requiring two years, so distinguish mandatory requirements from weighted preferences. Your feasibility plan should document dataset permissions, explain privacy boundaries for local versus hosted deployment, and estimate operating costs against the proposed $85 annual subscription.
- Claude Opus 5.5Preliminary57/100
- Strengths
- Your prototype goes beyond a chatbot wrapper. It is a deployed Streamlit and FastAPI system with TF-IDF similarity, skill coverage, what-if toggling of required competencies, and a ranked candidate matrix, as the prototype screenshot shows. Your choice of fully local inference, with blind screening of PII, evidence spans tied to resume sentences and regex negation handling ("no knowledge of Docker"), gives a credible privacy and explainability story for HR. Your submission honestly lists what broke: noisy CV formats, differently worded skills, imbalanced labels, and similarity scores that did not reflect real suitability.
- To improve
- Your materials contradict each other on the core model: the submission says Logistic Regression and the README says Random Forest. You also never explain how match labels were created for UpdatedResumeDataSet, which only has job categories. That makes the 91% accuracy on 100 samples hard to trust. The "industry ATS baseline ~65%" has no source, so replace it with a real comparison: time a recruiter shortlisting the same CVs by hand against your tool, and show concrete misranked candidates. Name one specific user and one measured outcome, such as minutes saved per 50 CVs or agreement with recruiter shortlists, instead of the illustrative -42% time-to-hire figures. Finally, sanity-check the $85/year price against your hosting and support costs.
The team
Telman Vəliyev
designer
Nadir QədirzadəLead
data
Fəqan Qafarov
developer