Zhane's Portfolio Vercel is the public command center for Zhane Grey's Vercel-hosted AI engineering portfolio. The hub keeps Build Doctor as the flagship deploy-debugging tool while linking the broader portfolio surface: resume tailoring, evidence dashboards, gateway failover, enterprise agent workflows, RAG/digital twin systems, visual systems, DatumX, and supporting Vercel demos.
Build Doctor is a Vercel-deployed developer tool that turns failed Next.js and Vercel build logs into a traceable root-cause diagnosis, safe patch draft, optional DeepSeek review, and exportable markdown incident report.
The deterministic engine is the source of truth. Optional LLM review is isolated behind a fail-closed enrichment layer and only receives sanitized diagnosis data.
- Portfolio hub: https://vercel-build-doctor-agent.vercel.app
- Build Doctor: https://vercel-build-doctor-agent.vercel.app/build-doctor
- Case study: https://vercel-build-doctor-agent.vercel.app/case-study
- Reviewer packet:
REVIEWER_PACKET.md - Engineering log:
ai-engineering/daily-engineering-log.md - Build Doctor docs:
docs/build-doctor.md - Release evidence:
docs/build-doctor-release-evidence.md
- Vercel Build Doctor Agent - GitHub
- AI Resume Tailor Service - GitHub
- Zhane Grey Evidence Dashboard - GitHub
- AI Gateway Failover Playground - GitHub
- Enterprise Agent Workflow Studio - GitHub
- Agentic RAG Memory Digital Twin Edge System - GitHub
- Fuji by ZRT - GitHub
- Resume Evidence RAG Auditor
- DatumX - GitHub
- Untitled X Nine
- UI Pi Eight
This close-up frame keeps the README atmospheric without weakening the technical layout.
Build Doctor supports a five-step workflow:
- Paste build logs
- Diagnose root cause
- Show local trace
- Suggest patch draft
- Export report
The app:
- Redacts secrets before display, reporting, or optional provider review.
- Classifies the likely failure with deterministic rules.
- Extracts evidence lines and affected subsystem signals.
- Builds a local diagnostic trace and safe patch draft.
- Offers deterministic suggested solutions and an editable fix plan.
- Optionally requests a DeepSeek review through OpenRouter using sanitized diagnosis JSON only.
- Exports a markdown incident report with evidence and remaining risks.
Build Doctor is not a hallucination-first debugging agent.
The local pipeline owns:
- redaction
- failure classification
- evidence extraction
- diagnostic trace
- patch draft
- suggested solutions
- markdown export
The optional DeepSeek layer can improve the explanation, but it does not replace deterministic diagnosis and it does not apply code changes.
Build log input
-> redact secrets
-> classify failure
-> extract evidence
-> map remediation
-> build patch draft
-> build suggested solutions
-> export markdown report
-> optional sanitized DeepSeek review
Primary implementation areas:
src/lib/redact-secrets.tssrc/lib/log-parser.tssrc/lib/failure-taxonomy.tssrc/lib/patch-recipes.tssrc/lib/build-doctor/index.tssrc/lib/build-doctor/openrouter.tssrc/lib/build-doctor/llm/openrouter-deepseek.tssrc/lib/build-doctor/solution-suggestions.tssrc/app/api/diagnose/route.tssrc/app/api/enrich/route.tssrc/app/api/report/route.ts
Current deterministic taxonomy includes:
MISSING_ENV_VARTYPESCRIPT_ERRORMODULE_NOT_FOUNDNEXT_BUILD_ERRORNEXT_STATIC_GENERATION_ERRORAPP_ROUTER_ROUTE_HANDLER_ERRORPACKAGE_INSTALL_ERRORPACKAGE_JSON_PARSEPNPM_LOCKFILE_MISMATCHSPAWN_PERMISSIONPRISMA_DATABASE_ERRORSUPABASE_CONFIG_ERRORSTRIPE_WEBHOOK_ERRORVERCEL_ENV_VAR_MISSINGVERCEL_RUNTIME_ERRORSERVERLESS_FUNCTION_LIMITOUT_OF_MEMORYESLINT_BUILD_ERRORVITE_BUILD_ERRORUNKNOWN
OpenRouter enrichment is optional and guarded.
Current live model target:
deepseek/deepseek-v4-flash:free
Safety rules:
- paid models are blocked by default
openrouter/freeis blockedopenrouter/autois blocked- raw logs are not sent to the provider
- secrets are not sent to the provider
- plain JSON output is parsed and Zod-validated
- invalid provider output fails closed
- deterministic diagnosis remains available when provider review fails
Step 4 includes deterministic suggested solutions for major failure classes.
The user can:
- review solution cards
- inspect likely affected files
- copy snippets
- copy verification commands
- autofill an editable fix plan
- include selected solutions in the exported report
These suggestions do not auto-apply patches and do not require a model provider.
Security controls currently verified in the repo:
- secret redaction before UI output
- secret redaction before report export
- secret redaction before optional provider review
- no paid-model fallback by default
- read-only
GET /api/integration-health - public integration-health write probes removed
.vercelignoreexcludes local.env,.omx,.next,node_modules, and test artifacts from deployment upload- Vercel production env vars are configured as encrypted project variables
Supabase status:
- project:
gajpnqqfkjtmqdnufbcf - evidence tables present:
suite_events,demo_runs,eval_runs,exported_reports - probe-like rows cleaned from
suite_events
Latest verified release checks:
npm run typechecknpm run testnpm run buildnpm run test:e2enpm audit --audit-level=moderatenpm run audit:securitynpm run audit:45knpm run audit:report
Latest verified results:
- unit tests:
25passing - E2E tests:
4passing - npm audit:
0 vulnerabilities - security audit:
4,000 / 4,000 - premium audit:
45,000 / 45,000
Generated audit artifacts:
The repo includes a running verified engineering log:
Use it as the canonical proof trail for:
- shipped changes
- verification commands
- deployment evidence
- resume-safe summaries
Recent verified themes from the log:
- deterministic Build Doctor architecture and failure taxonomy expansion
- DeepSeek free OpenRouter integration with fail-closed guards
- employer-facing UI and report copy polish
- deterministic suggested solutions and editable fix plans
- Supabase readiness verification and probe cleanup
- release hardening, dependency vulnerability remediation, Vercel deployment hardening, and GitHub publication
Key paths:
src/app/build-doctorUI routes and page entry pointssrc/components/Build Doctor UI componentssrc/lib/build-doctor/diagnosis and provider orchestrationsrc/test/build-doctor.test.tsdeterministic unit coveragee2e/build-doctor.spec.tsPlaywright browser coveragedocs/deeper product and release documentationscreenshots/visual QA evidenceai-engineering/verified engineering log
npm install
npm run devOpen:
http://localhost:3000/http://localhost:3000/build-doctor
Useful commands:
npm run typecheck
npm run test
npm run build
npm run test:e2e
npm run audit:security
npm run audit:45k
npm run audit:reportDeterministic mode works without paid provider access.
See:
Important variables:
ENABLE_LLM_ENRICHMENT=false
LLM_PROVIDER=mock
OPENROUTER_API_KEY=
OPENROUTER_MODEL=deepseek/deepseek-v4-flash:free
OPENROUTER_SITE_URL=http://localhost:3000
OPENROUTER_APP_TITLE=Build Doctor
ALLOW_PAID_LLM_MODELS=false
NEXT_PUBLIC_SUPABASE_URL=
NEXT_PUBLIC_SUPABASE_ANON_KEY=
SUPABASE_SERVICE_ROLE_KEY=Production deploy target:
Latest verified production deployment:
- deployment id:
dpl_251nHCbKjvck2zthFXunuCcTEG2p
GitHub repository:
Latest release commits:
9160af1Harden Vercel deployment inputsdf04d48Document Build Doctor release evidencecfb9b52Release Build Doctor security hardening
Selected visual evidence:
screenshots/build-doctor-final-desktop.pngscreenshots/build-doctor-final-mobile.pngscreenshots/build-doctor-final-diagnosed.pngscreenshots/build-doctor-suggested-solutions.png
- Diagnosis is deterministic and log-based, not live repo introspection.
- Optional provider review can be rate-limited.
- Suggested solutions are template-driven and should be reviewed by a developer before code changes.
- This is a reviewer-facing engineering product, not a multi-tenant SaaS platform.
- Built Build Doctor, a Vercel-deployed developer tool that diagnoses failed Next.js and Vercel build logs through a deterministic taxonomy, redacts secrets, generates traceable root-cause receipts, suggests safe patch drafts and deterministic remediation plans, optionally enriches explanations through DeepSeek via OpenRouter, and validates reliability with unit, browser, and audit coverage.
Core XRPL EVM systems plus related public product and AI repositories from the same portfolio.
| Project | Description | Status |
|---|---|---|
| ZUC Mine Command Center | On-chain uranium mining operations dashboard with real-time reserve tracking, miner registry, and direct contract interaction through a frontend-only control surface. | Live |
| U235 Fuel Cycle | Deterministic XRPL EVM fuel-cycle pipeline that tracks uranium batches from ore to enriched fuel rod with full on-chain traceability. | Live |
| ISR Network | In-situ recovery control system with on-chain asset tracking, lifecycle state transitions, and operator-facing industrial simulation. | Live |
| Dark Matter Farm | XRPL EVM staking protocol with three orbit tiers, lock-period yield mechanics, and event-driven reward emissions. | Live |
| Cohr Lab | Semiconductor laser fabrication lifecycle modeled as an immutable on-chain state machine from crystal growth to final pigtail. | Live |
| ForgeX | Foundry-powered XRPL EVM deployment console that combines a natural-language UI, Node CLI orchestration, and realtime shader-based visuals. | Live |
| DatumX | Verification protocol for AI-transformed industrial data with deterministic lineage, validator review, and XRPL EVM finalization. | Live |
| Ethex Lottery Game | Foundry plus Next.js betting workflow that modernizes the EthexLoto lifecycle for XRPL EVM reviewer-facing execution. | Public Repo |
| 3DMoonX | Cinematic lunar industrial-base experience that combines Blender source assets with a React Three Fiber web runtime. | Live |
| Unknown002 | Browser-based 3D engineering viewer for a nuclear-electric propulsion spacecraft concept with staged prompt-pack support. | Public Repo |
| AI Engineering Evidence Engine | Interactive evidence dashboard that turns local engineering proof into a reviewer-facing systems narrative. | Live |
| Build Doctor | Codex-style build diagnosis harness for failed Next.js and Vercel builds with deterministic failure analysis. | Live |
| AI Gateway Failover Playground | Public-facing sandbox for request routing, provider fallback, and resilient AI gateway behavior. | Live |
| Enterprise Agent Workflow Studio | Public-facing studio for approval-gated enterprise agent workflows, risk scoring, and audit-oriented design. | Live |
| Resume Evidence RAG Auditor | Public-facing proof surface for claim verification, evidence retrieval, and grounded resume bullet generation. | Live |
| AI Resume Tailor Service | Static Vercel-ready application for evidence-backed resume, cover-letter, and job-packet tailoring. | Live |
| Fuji | Cinematic Next.js Fuji gallery atlas for portfolio storytelling and visual system design. | Live |
| AI Agents for Beginners | Lesson repository for getting started building AI agents. | Public Repo |
| Agentic RAG Memory Digital Twin Edge System | Public-facing landing page for an agentic RAG, memory, and digital-twin edge-system portfolio project. | Live |