PDFs. Hateful, inconsistent little monsters. He’s staring at the screen, mammoth.js choking on some scanned-in-Word relic, text extraction failing spectacularly.
That’s where ResumeeNow began its slide from ‘quick weekend project’ to full-blown AI-powered resume platform. Import a resume? Sure. Tailor it with AI? Why not. ATS audit, cover letter gen, flawless PDF out? Hold my beer.
But here’s the thing. What sounded like forms plus a button turned into a beast. Connected systems. All agreeing. Or else.
The dev thought CRUD. Reality? Import feeds builder feeds AI feeds export. One mismatch, and users rage-quit.
Why Did a Resume Builder Need Military-Grade Architecture?
Simple apps forgive slop. Loose state. One-way flows. Not this.
Imported data had to match AI tweaks, which had to match the exported PDF—looking sharp on any printer, not just screen.
He calls it a platform now, not a builder. Smart rebrand. Because engineering constraints flipped. No more happy-path assumptions.
“The hard part was not just building screens. The hard part was making import, editing, AI workflows, export, notifications, and account-aware behavior feel like one coherent app.”
Spot on. That’s the quote that nails it. Screens are easy. Coherence? Blood, sweat, server bills.
His fix: Boundaries. Ruthless ones.
Public landing. Auth’d dashboard. Builder. Admin gates. Print paths. Each in its own route silo.
} />
} />
}>
} />
} />
Genius. Or sanity-saving, at least. No monolith shell leaking assumptions everywhere. Print stays isolated—no auth leaks, no client bloat. Admin? Double-gated. Public? Pure, untainted.
This router split? It aged like fine wine. When backend piled on—AI gateways, Supabase edges—it slotted right in. Client stays lean; servers own the risky bits.
Most apps ignore this. End up as tangled frontends. ResumeeNow dodged that bullet early.
Is PDF Import the Devil’s Work?
Hell yes.
Empty fields? Trivial editor. Upload PDF? Pipeline time.
Auth check. Validate upload. Server-side text rip. Usability sniff. Then parse to structured gold.
“if (!modules.isReadableDocumentText(joined)) { throw new Error(‘Could not extract reliable text from this PDF. This often happens with scanned or image-based PDFs without OCR.’); }”
Love this. Fail fast, tell why. No half-parsed garbage poisoning the builder.
Scanned PDFs? Image-based trash? Gatekeep ‘em. Better a clear ‘nope’ than editable lies.
Tools: mammoth, pdfjs-dist, unpdf. Playwright for browser-rendered PDFs—because server fonts hate you.
Result? Normalized resume shape. Sections clean, editable, AI-ready.
But wait—ATS audits? AI tailoring? That’s next-level glue.
Stack time. React 19, Vite 7, TS, Tailwind 4, React Router 7, Query 5, Zustand 5. Supabase for auth/Postgres. Vercel APIs. Edge functions.
Move-fast stack. But only ‘cause he sliced responsibilities thin.
Frontend: Feature layers. Routes, builders, dashboards.
Backend: Privileged APIs. AI via gateway—no client scatter. Exports server-side.
Four user surfaces. Two server runtimes. Evolved from needs, not buzz.
Skeptical take? Supabase+ Vercel—hot, but lock-in risk. Edge functions shine for AI bursts, though. Bills could spike.
Why Does Router Separation Beat Fancy State Managers?
Zustand’s great. React Query too. But routes as moats? Underrated.
No shared app shell dragging public into auth hell. Print paths unpolluted. Admin isolated.
Later adds—like notifications—piggyback easy. Clear contracts: Client asks, server delivers.
Compare to typical React sprawl. Giant Context. Leaky providers. Debugging nightmare.
ResumeeNow’s way scales. Add surfaces? New route. Done.
Dry humor: If only more devs treated routers like firewalls, not afterthoughts.
Four pillars hold it up.
-
Import pipeline—harsh but fair.
-
Normalized data shape—AI’s happy place.
-
AI gateway—rate-limits, keys safe.
-
Export ritual—Playwright renders pixel-perfect.
ATS audit? Scans keywords, scores. AI tailor? Role-specific tweaks. Cover letters? Prompt magic.
All synced. Edit ripples everywhere. Magic—until AI hallucinates.
Can This Stack Handle Real Scale?
Short answer: Maybe.
Supabase Postgres scales. Vercel edges too. But AI calls? OpenAI bills await.
Playwright for PDFs? Resource hog. Optimize or die.
Unique insight: This echoes 2010s resume mills—PDF woes killed ‘em. AI revives it, but commoditization looms. LinkedIn’s baking this in; ResumeeNow’s indie charm won’t last. Bold prediction: Acquired by 2026, or cloned by VCs.
PR spin? None here—straight engineering confessional. Refreshing. No ‘revolutionary’ fluff.
Corporate hype elsewhere calls AI builders ‘future of hiring.’ Please. Resumes die slow; ATS gatekeep forever.
Users love it. Import works. Tailors beat manual. Exports crisp.
But watch notifications—Supabase realtime? Slick, if not abused.
Admin dashboard? Power users only.
Overall? Solid evolution. From naive forms to coherent beast.
Lessons: Boundaries first. Fail parsing loud. Normalize ruthlessly.
🧬 Related Insights
- Read more: Chainguard EmeritOSS Swoops In to Save MinIO and a Fleet of Forgotten Open Source Heroes
- Read more: crossword.by: One Dev Ditches Ad Hell for Blazing-Fast, Ad-Free Puzzles in 7 Languages
Frequently Asked Questions
How does ResumeeNow parse uploaded resumes?
Server-side extraction with mammoth/pdfjs, text cleanup, readability check—rejects scans politely.
What’s the tech stack behind ResumeeNow?
React 19/Vite/TS/Tailwind, Supabase/Vercel, Zod validation, Playwright for PDFs.
Will AI resume tools like ResumeeNow replace manual editing?
Not yet—great starters, but humans tweak for nuance. AI hallucinates edge cases.
