CandidAI

Run locally in a virtual environment

Uses the following dependencies scikit-learn, anthropic, streamlit, PyPDF2, dotenv, voyageai, numpy

Inspiration

Recruitment bias still limits opportunities for women and underrepresented groups in tech. We’ve all seen great candidates passed over because of names, gender cues, or where they studied, not because of their skills. We wanted to build a tool that challenges that bias directly, helping employers focus on what truly matters: ability, experience, and potential.

What it does

CandidAI automatically redacts personal identifiers from resumes and evaluates how well a candidate’s skills align with a job description. It removes gendered terms, university names, and other demographic clues, creating a fairer first impression. Then it uses natural-language processing and embeddings to compare technical experience, education, and projects with job requirements, offering an ethical, data-driven match score.

How we built it

We used Streamlit for the interface, Anthropic’s Claude for iterative redaction and bias verification, and VoyageAI embeddings for similarity scoring. The system cycles between “debias” and “verify” steps until the text is clean, then computes alignment between anonymized resumes and job postings.

Challenges we ran into

Balancing redaction accuracy with context preservation was difficult, removing names and pronouns without erasing valuable details. Integrating multiple LLM tools while maintaining a consistent feedback loop also required careful state management.

Accomplishments that we're proud of

We built a full end-to-end redaction and evaluation system that actually works. Our app can anonymize resumes in seconds, highlight fairness as a design principle, and demonstrate how AI can be used for equity rather than exclusion

What we learned

We learned how to chain LLM tools effectively, how small UX details (like transparency and clear feedback) build trust, and how bias can hide in unexpected linguistic patterns.

What's next for CandidAI

Next, we want to expand the platform into a bias-audit toolkit for HR teams, integrating dashboards, explainability features, and multilingual resume support. We also hope to partner with diversity-focused organizations to make inclusive hiring technology accessible to everyone.

Built With

• agents
• anthropic
• python
• streamlit