CredAI: Turn skills into proof

Turn resume skills into proof using AI-evaluated exams and verifiable on-chain credentials.

Comment

CredAI: Turn Skills into Proof

Inspiration

Resumes are built on trust, but trust is exactly what's missing.

During internships, hackathons, and recruiting cycles, I kept seeing the same problem: candidates list skills like Python debugging or SQL on their resumes, but recruiters have no reliable way to verify them without long interviews or take-home tests. Certifications often test memorization, not real understanding, and are easy to inflate or fake.

I wanted to flip the model.

Instead of claiming skills, what if candidates could prove them; quickly, fairly, and in a way that anyone could independently verify?

That question became CredAI.

What it does

CredAI turns skills into proof.

CredAI is an AI-powered skill certification system that evaluates real competence through a short, interactive oral exam and issues a verifiable micro-credential backed by cryptographic proof.

A candidate:

  • Enters any skill (for example, Python debugging)
  • Takes an AI-led oral exam with practical reasoning questions
  • Receives a pass, fail, or review verdict based on a transparent rubric
  • If they pass, gets a credential whose evidence hash is anchored on Solana

A recruiter:

  • Pastes the credential ID into a verification page
  • Instantly verifies authenticity, score summary, and proof on-chain
  • No trust in the issuer required

How I built it

I designed CredAI end-to-end as a real assessment pipeline, not a quiz app.

Architecture overview

Frontend

  • Built with Next.js (App Router) and Tailwind CSS
  • Candidate exam flow, recruiter verification page, and admin dashboard

Backend

  • FastAPI (Python) for session orchestration
  • Google Gemini 2.5 Flash for:
    • Generating practical, skill-specific questions
    • Grading answers using rubric-grounded evaluation
    • Asking adaptive follow-up questions

Assessment logic

  • Each exam is scored using a numeric rubric (0–10)
  • Follow-up questions adapt based on reasoning quality
  • Final verdict: PASS, FAIL, or REVIEW

Verification layer

  • Exam evidence is serialized into canonical JSON
  • A SHA-256 hash is computed:

$$h = \text{SHA-256}(\text{exam_evidence})$$

  • The hash is written to Solana devnet via a Memo transaction
  • The resulting transaction signature becomes the verification anchor

Admin dashboard

  • Session analytics
  • Full transcripts and scores
  • Blockchain proof links for auditing

Challenges I ran into

Designing fair AI evaluation

Grading reasoning (not just correctness) required careful prompt design and rubric grounding to avoid shallow answers passing.

Preventing "quiz-style" shortcuts

I intentionally avoided multiple-choice questions and forced explanations and debugging steps to ensure depth.

Balancing speed vs rigor

The exam had to be short enough for real use, but deep enough to be meaningful. I solved this with adaptive follow-ups instead of long tests.

Making blockchain useful (not gimmicky)

I focused on blockchain as an audit anchor, not as a data store, keeping sensitive data off-chain while preserving verifiability.

Accomplishments that I am proud of

  • Built a working, end-to-end skill certification system, not just a demo
  • Designed an AI oral exam that adapts to arbitrary skills entered as free text
  • Implemented cryptographically verifiable credentials that anyone can check
  • Created a transparent admin dashboard showing full evaluation evidence
  • Delivered a project that addresses a real, global hiring problem

What I learnt

  • Skill assessment is as much about reasoning transparency as accuracy
  • AI systems are far more trustworthy when their decisions are auditable
  • Blockchain adds real value when used for immutability and accountability
  • Clear problem framing matters as much as technical sophistication

Most importantly, I learned that trust itself can be engineered.

What's next for CredAI

  • Persistent storage (SQLite or Firebase)
  • PDF certificates with QR-code verification
  • Authentication for recruiters and issuers
  • Skill history and multiple attempts per candidate
  • On-chain credential registry for long-term verification
  • Expanded skill rubrics beyond programming

CredAI doesn't replace resumes.
It makes them honest.

By turning skills from unverifiable text into auditable proof, CredAI makes hiring fairer, faster, and more transparent.

Built With

  • agentic-ai
  • fastapi
  • google-gemini-2.5-flash
  • google-genai-sdk
  • javascript
  • next.js
  • pydantic
  • python
  • react
  • restful-api-design
  • sha-256-hashing
  • solana-devnet
  • solana-memo-program
  • solana-python-sdk
  • tailwind
Share this project:

Updates