ApplyProof

Inspiration

International students often apply to hundreds of jobs without knowing whether an employer can actually sponsor them. Job posts are vague, company names do not always match legal sponsor names, and students can waste time applying to roles that were never realistic.

ApplyProof was built to make that decision clearer. Instead of giving generic career advice, it verifies an opportunity against official evidence and gives the student a practical next step: Apply, Verify, or Skip.

What It Does

ApplyProof is an evidence-first AI workflow for checking student job opportunities.

A user can paste a vacancy or enter an employer name. ApplyProof then:

• extracts the likely employer from the job text
• checks it against the official GOV.UK sponsor register
• compares the student’s preferred visa route against the sponsor licence
• explains the reasoning through an agent trace
• gives a clear decision: Apply, Verify, or Skip
• provides practical next actions for the student

The product currently indexes 126,973 official organisations and 141,425 licence rows from a real GOV.UK sponsor register snapshot.

How We Built It

We built ApplyProof as a real web app using Next.js, React, TypeScript, Tailwind CSS, and shadcn/ui inside a PulseUI dashboard template.

The verification engine runs through an agent-style pipeline:

1. Context agent prepares the opportunity brief.
2. Extraction agent identifies employer candidates from the vacancy.
3. Evidence agent matches candidates against official sponsor data.
4. Decision agent scores the match and route coverage.
5. Action agent turns the result into next steps.

The sponsor data is loaded from a compact official register dataset and indexed in the browser for fast matching. The UI preserves the existing dashboard layout while replacing the generic dashboard content with a working ApplyProof product experience.

Challenges We Faced

The hardest challenge was avoiding a fake AI demo. It would have been easy to build a chatbot that simply sounded confident, but that would not solve the real problem.

Instead, we focused on evidence. Employer names can be messy, shortened, or written differently from official legal names, so we had to build normalization, candidate extraction, fuzzy matching, route checking, and confidence scoring. We also had to make the reasoning visible so users can inspect why the product made a decision.

Another challenge was time. The hackathon deadline forced us to prioritize the core workflow: real data, real matching, clear decisions, and a demo that proves the product works end to end.

What We Learned

We learned that useful AI products do not always need to hide complexity. For high-stakes decisions, users need to see the evidence, not just the answer.

We also learned that agentic development works best when each agent has a narrow responsibility. ApplyProof became stronger when we separated extraction, evidence matching, decisioning, and action planning instead of treating everything as one black-box response.

What’s Next

Next, ApplyProof could add live register refreshes, browser extensions for LinkedIn and job boards, user accounts for saved opportunities, and an LLM extraction layer for more complex job descriptions while keeping official sponsor data as the source of truth.

Built With

• agentic
• ai
• css
• data
• evidence
• ffmpeg
• gov.uk
• matching
• next.js
• node.js
• playwright
• pulseui
• react
• register
• retrieval
• shadcn/ui
• sponsor
• tailwind
• typescript