Models duplicate delivery as a first-class failure mode.
High confidenceThe consumer stores processed event ids before side effects and the tests replay the same payload twice to prove convergence.
maya/event-ledger · src/replay/consumer.tsSignal without the black box.
This is what Truen is meant to feel like: a hiring report with restraint, visible uncertainty, and concrete evidence under every major claim.
Read-only Truen Profile
Maya Chen
Founding full-stack engineer · Toronto, CA · Generated Mar 14, 2026
Overall signal
Worth interviewing. Strong systems reasoning; verify ownership live.
Maya shows stronger-than-typical evidence for early-career systems thinking: she decomposes failure modes, writes validation around edge cases, and revises AI-assisted code before it lands. The profile does not claim production impact beyond observable repository signals.
Evidence inspected
maya/event-ledger
318 authored commits, queue replay service, CI and integration tests detected.
maya/clinic-scheduler
Multi-role app with auth flows, migrations, and documented scheduling tradeoffs.
Trajectory
Accelerating
Later commits show more pre-code planning, smaller changes, and better tests than the first project slice.
Systems thinking
Strong
Architecture notes and code structure both show explicit treatment of replay, idempotency, and latency. The strongest signal is that the design changed after failure-mode testing rather than staying at scaffold level.
Open evidence
Documents tradeoffs before implementation hardened.
High confidenceDesign notes compare latency cost against deterministic recovery and call out why the first version rejected optimistic acking.
maya/event-ledger · docs/replay-design.mdProblem-solving depth
Strong, with focused scope
The repository history shows a narrow problem worked through multiple layers: schema, queue boundaries, tests, and operational notes. It is not broad platform work, but the depth inside the chosen slice is credible.
Open evidence
Debugging path is visible in commit sequence.
High confidenceA failed duplicate-delivery test led to cursor movement being separated from handler completion, then covered by a regression test.
commit pattern · "fix replay cursor advancing on failed delivery"Avoids hiding unknowns.
Moderate confidenceThe limitations doc names untested broker behavior and proposes a manual verification checklist instead of claiming end-to-end production readiness.
maya/event-ledger · docs/limits.mdCode quality
Moderate to strong
The codebase is readable and consistently typed. Error handling is strongest in the event pipeline and weaker in admin surfaces, so this should be probed rather than treated as uniformly mature.
Open evidence
Validation is close to external boundaries.
Moderate confidenceForm inputs pass through explicit schemas before database writes, with user-facing errors separated from server logs.
maya/clinic-scheduler · src/server/actions.tsSome peripheral UI paths remain conventional.
Low confidenceAdmin tables are functional but lightly abstracted; there is less evidence of thoughtful empty/error state handling here.
maya/clinic-scheduler · app/admin/page.tsxAI collaboration
Enhanced, not delegated
Available traces suggest AI was used for acceleration, then edited against the candidate’s own constraints. Confidence stays moderate because tool telemetry is not connected in this sample.
Open evidence
AI output was revised before acceptance.
Moderate confidenceThe follow-up commit removes broad generated branching and replaces it with a smaller reducer that matches the replay invariant.
commit pattern · "replace generated merge path with deterministic reducer"Tool-level usage is not directly verified.
Insufficient evidenceThis sample infers collaboration style from repository history only; connected AI tool telemetry would be required for higher confidence.
profile limitation · AI tools disconnectedImpact proxies
- Deployment URL detected for clinic scheduler, but no production usage metrics are claimed.
- Issue history shows two externally reported bugs closed with follow-up tests.
- Instrumentation config is absent, lowering confidence in runtime maturity.
Risks to probe
- Needs live probing on ownership of AI-assisted refactors.
- Admin and reporting surfaces show less resilience than core pipelines.
- Most collaboration evidence comes from solo work, not team review.
Limitations
- This profile is based on selected repositories only.
- No private production logs, employer references, or live pair-programming evidence were used.
- AI collaboration confidence remains capped without connected tool signals.
Interview follow-up
Walk through the replay cursor bug and explain what invariant the final design protects.
Show one AI-assisted change you rejected or rewrote, and explain the difference.