SkillVerify AI

Inspiration

Generative AI tools promise to help job seekers tailor resumes in seconds, but standard LLMs frequently hallucinate skills, tools, or years of experience that candidates do not actually possess. When candidates present these fabricated details in job applications, it leads to failed technical screenings and damaged professional credibility.

What it does

SkillVerify AI compares a candidate’s Master Resume against a Target Job Description and delivers a verified match report:

Grounded Summary Synthesis: Generates a tailored professional summary strictly bounded by the candidate's actual experience.

Deterministic Verification Audit: Runs a post-generation verification script (verifier.py) that audits every claimed technology against the raw Master Resume text.

Smart Skill Reclassification: Displays 100% verified matched skills while cleanly categorizing missing job requirements into a dedicated list.

Visual Guardrails: Highlights verified claims with green grounding badges and isolates unverified terms.

How we built it

Backend: Python and Flask web framework.

Local AI Engine: Integrated with Ollama running llama3.2:1b on-device for 100% data privacy.

Verification Layer: Deterministic string matching and set-difference audit logic in Python (verifier.py).

Frontend: Jinja2 HTML5 and CSS templates with dynamic badge rendering

Challenges we ran into

Lightweight local LLMs often copy required skills from the job description directly into the "Matched Skills" list, even when missing from the candidate's resume. Prompt engineering alone could not prevent this. We solved it by building a secondary deterministic verification layer in Python that inspects the raw resume text and automatically reclassifies ungrounded skills into "Missing Job Requirements."

Accomplishments that we're proud of

Built a completely local, zero-subscription tool that ensures full candidate data privacy.

Implemented a hybrid architecture combining generative text synthesis with hard-coded software verification.

Successfully eliminated false positives, achieving a reliable 100% Grounded verification workflow.

What we learned

AI language models excel at summarizing and formatting text, but software validation logic must remain the ultimate authority on factual truth in high-stakes domains like hiring.