ShieldHire AI | Devpost

💡 Inspiration

Traditional technical recruitment suffers from a massive catch-22. Companies need verified proof of engineering talent, but storing unencrypted, raw candidate credentials on public databases introduces severe data-leak liabilities. At the same time, unconscious human bias (based on names, gender, or locations) consistently corrupts the initial resume screening process. We built ShieldHire AI to securely decouple a developer's real-world identity from their true technical merit.

⚙️ What it does

ShieldHire AI handles data processing in a unique three-layer pipeline:

The Objective AI Filter: A recruiter pastes a candidate's raw resume text into the system. The Google Gemini API instantly sanitizes the text, completely stripping away Personally Identifiable Information (PII) to eliminate bias while calculating a pure technical merit score based strictly on experience.
The 1AM Wallet Handshake: The candidate authenticates securely using the 1AM browser extension, verifying their cryptographic signature without revealing real-world registration details.
The Zero-Knowledge Vault: The merit score is fed into a custom Midnight smart contract (shield_hire.compact). A cryptographic proof is generated to prove the candidate meets or exceeds the company's hiring threshold (Score >= 70) without exposing the candidate's name or revealing their exact score to the public ledger.

🛠️ How we built it

Frontend Architecture: Built using React and Vite, styled with a custom dark-engine layout optimized for high-end enterprise scannability. It interacts with the 1AM extension by capturing and responding to global injected window state objects (window.midnight).
Backend Processing: An Express server handling secure request-routing to the Google Gemini API.
Smart Contract Layer: Written in the Compact smart contract language and compiled via WSL using a standalone target-specific Linux musl toolchain to create the underlying cryptographic validation circuits.

💥 Challenges we ran into

Integrating a bleeding-edge, mainnet-live privacy layer like Midnight with generative AI models presented distinct architectural challenges. We had to navigate early-stage extension naming conventions in browser memory allocation to avoid component unmounting states during wallet approvals, and we had to establish defensive data wrappers on the client to safely parse raw cryptographic returned types without throwing UI errors.

🏆 Accomplishments that we're proud of

We successfully designed and executed a highly technical, end-to-end user loop. We didn't just build a visual wrapper—we compiled operational ZK circuit logic, connected a live extension wallet to an interactive app state, and cleanly implemented an automated generative AI redaction schema all within a 48-hour build window.

🎓 What we learned

We gained profound hands-on experience with production-ready decentralized applications (dApps). We learned how to model business constraints inside zero-knowledge assertions, compile smart contracts targeting live networks using isolated compiler binaries, and orchestrate private states against public ledger footprints.

🔮 What's next for ShieldHire AI

We plan to scale ShieldHire AI by introducing fully tokenized, automated ZK-Credentials. Candidates will be able to store their verified skills permanently inside decentralized storage, allowing recruiters to purchase anonymous interview access keys directly via confidential dApp transactions, creating a completely zero-bias marketplace for global talent.