-
Open the original work with the exact quotation highlighted.
-
Recall the meeting, then inspect passages relevant to a role.
-
Maintain the profile, project notes, and files shared with connections.
-
Keep specific personal context for each recruiter connection.
Inspiration
A recruiter can leave a career fair remembering a promising conversation, then lose that context in an inbox full of applications. A connection request preserves a name; the reason that person stood out can be harder to recover.
We chose a narrow part of Side A: Trust coming in, the applicant flood: help recruiters revisit people they already met. LinkedIn already supports QR connections, recruiter notes, and candidate rediscovery. Our focus is bringing a candidate-authored encounter record and their maintained work together, with passages the recruiter can inspect against a role.
What it does
The recruiter shares an event QR. The candidate records a memorable detail and what they discussed, then maintains their profile, project notes, and uploaded work. The recruiter is not asked to write hundreds of recaps. The original recap remains available after edits.
Later, the recruiter opens that person, recalls the conversation, and selects a role. Relevant passages appear beside the meeting note; opening a source highlights the exact quotation. Email and supplied profile links let the recruiter resume contact themselves. There are no candidate scores or automated outreach. Connections can be removed, and CSV export supports a manual handoff.
How we built it
We used Codex for implementation, code review, and test development, then checked the flows in the running desktop app.
The interface uses React, TypeScript, shadcn/ui, and Tailwind. Express handles sessions, access checks, QR creation, and PDF/text extraction. SQLite stores the records.
A small Rust executable performs lexical passage lookup. Optional GLM-5.3 through OpenCode Go organizes role context from bounded excerpts. The server validates source IDs and exact quotations, caches valid results, and falls back to Rust when model responses fail validation or the provider is unavailable. There is no vector database or separate retrieval service.
Challenges we ran into
The difficult part was keeping context dependable as people edit work or switch accounts. We added access checks, cancellation and account guards, retained original recaps, and refreshed role context when its inputs changed. Another challenge was keeping the recruiter view focused on the person while making source inspection available beside it.
Accomplishments
The desktop walkthrough connects both sides: invitation, meeting note, maintained work, role context, exact source, and external contact. Verification includes 69 Node tests and 11 Rust tests. A live GLM-5.3 check returned valid quotations and confirmed cache reuse. File replacement preserves the existing document if validation or saving fails.
What we learned
Moving recap entry to the candidate changes who does the work; it does not eliminate effort. A source quotation makes an explanation inspectable, but does not verify the claim. A QR grants invitation access, not proof of attendance, identity, or endorsement. We have not measured time savings or hiring outcomes.
What's next
Test whether recruiters can recall the right encounter and find supporting work after a delay. Measure candidate completion effort too. Further work includes production account recovery and operational safeguards. ATS integration is not implemented; current export is CSV.
Team
- Prashant Shah: Developer
- Daniel Rajakumar: Developer
Built With
- express.js
- glm-5.3
- node.js
- opencode-go
- react
- rust
- shadcn-ui
- sqlite
- tailwind-css
- typescript