MIXEDBREAD · FRONTEND ENGINEER · PREP DOSSIER 4-WEEK PLAN · v1.0 · 21.04.2026
MXBAI
Role · Frontend Engineer · Remote · Full-time

Ship a portfolio-grade proof you can build Mixedbread's playground — in 28 days.

This is not a generic frontend study plan. It's a targeted prep track built around the five gaps you flagged — Next.js App Router, Core Web Vitals, animation polish, AI/vector-search concepts, and frontend system design — with GreatFrontEnd as the spine and one killer portfolio project as the output.

28d
Total Runway
5
Gap Areas
1
Flagship Project
~12h/wk
Minimum Commit

Strip away the job-spec language. Mixedbread needs three things.

Every "Frontend Engineer" JD sounds the same. This one isn't. Read between the lines and the role clusters into three pillars — production Next.js fluency, taste-level UI craft, and enough AI literacy to not embarrass yourself in a playground demo. Your prep should hit all three.

Pillar 01 · Non-negotiable

Next.js App Router at production depth

Server Components, streaming, parallel + intercepting routes, Route Handlers, Server Actions, caching layers. They said "App Router preferred" — it's a filter.

Pillar 02 · The differentiator

Taste — design sense + animation polish

"Beautiful, performant interfaces" and "intuitive and delightful" appear twice. They've lost candidates here before. Micro-interactions, motion, typography hierarchy.

Pillar 03 · The bar raiser

AI/vector literacy (nice-to-have, but)

Playgrounds for embeddings + reranking is literally the job. Knowing what cosine similarity is and why rerankers exist puts you ahead of 80% of applicants.

Support 01

Core Web Vitals + perf

"Sub-second page loads" — they measure. You'll need Lighthouse ≥ 95 and real answers on LCP, INP, CLS.

Support 02

shadcn/ui + design system thinking

Composable component libraries. How to extend, theme, and maintain. Turborepo/pnpm workspace experience lands extra.

Support 03

Real-time + testing

SSE/WebSockets for streaming AI responses. Vitest + Playwright for the engineering-culture signal. Both nice-to-haves that become credibility markers.

You already have GFE. That's the backbone. Everything else layers on top.

GreatFrontEnd covers the classic frontend interview stack — coding, system design, quiz, UI fundamentals. Don't grind it randomly. Map its modules to your weeks so it compounds with the portfolio work instead of competing with it.

GFE

Daily rhythm: 30 min quiz questions before standup (trivia + rapid recall) + 60–90 min of one coding question or one system design study every other day. Weekends: do one full mock under timed conditions. Don't binge — consistency beats sprints for interview recall.

Week 1 · Foundations
JS Quiz + Coding Easy

Closures, `this`, promises, event loop, debounce/throttle, deep clone. Build muscle memory on the fundamentals they'll probe in round 1.

Week 2 · UI Components
Coding Medium

Autocomplete, modal, tabs, accordion, infinite scroll, data table. These are exactly the kind of components that live in a playground UI.

Week 3 · System Design
Frontend SD

Work through News Feed, Autocomplete, Pinterest. Then apply the same framework to "Design an AI search playground" — that's the likely question.

Week 4 · Mocks
Full Timed Rounds

Two full mock interviews. Coding medium + SD. Record yourself thinking out loud. Watch it back. This is where most people skip — don't.

Each week has one theme, three tracks, one deliverable.

Three tracks run in parallel every week: GFE (recall), learning (depth), and building (the flagship project). By end of Week 4, you have a working AI search playground, a perf-optimised deploy, and enough recall to handle a live coding round cold.

Week01

Next.js App Router — rebuild your mental model

Foundations · Routing · Rendering · Data
PRIORITY Next.js 15 React 19
Learning Track
  • Read Next.js docs end-to-end (App Router only) routing → rendering → caching → styling → optimizing
  • Watch Lee Robinson's "Next.js App Router in 100 minutes" then rewatch 2x at pause points
  • Deep-dive: Server Components vs Client Components mental model the 'use client' boundary is the #1 thing to nail
  • Study: Suspense, streaming, loading.tsx, error.tsx
GFE Track
  • Daily: 30 min JS quiz (closures, promises, event loop)
  • 2x coding easy/medium: debounce, throttle, flatten, curry
  • Weekend: 1 full timed coding mock
Build Track
  • Scaffold the flagship projectsee §04 — "AI Search Playground"
  • Next.js 15 + shadcn/ui + Tailwind + TypeScript strict
  • Set up Turborepo OR pnpm workspace
→ End-of-Week Deliverable Deployed "Hello World" on Vercel. Monorepo scaffold. One page routes correctly.
Week02

AI concepts + the playground build begins

Embeddings · Reranking · Streaming · UI Components
GAP FILL AI/ML SSE
Learning Track
  • Jay Alammar: "Illustrated Word2Vec" + "Illustrated Transformer" read twice. this is your vocabulary foundation
  • What are embeddings? Vectors, cosine similarity, dimensionality use Mixedbread's own blog posts — literal homework signal
  • What is reranking? Why does it exist on top of vector search?
  • Play with Mixedbread's API for 1 hour. Hit /embed and /rerank endpoints.
  • Streaming UI patterns: SSE vs WebSockets, useChat from AI SDK
GFE Track
  • Coding medium: autocomplete (debounced), modal, tabs
  • UI fundamentals module — accessibility, keyboard nav, ARIA
  • Weekend: build GFE's "type-ahead search" from scratch, no hints
Build Track
  • Playground: embeddings visualizer page input text → show vector → 2D projection (PCA/t-SNE)
  • Playground: reranker comparison page (query + docs in, ranked out)
  • Wire up streaming responses via SSE
→ End-of-Week Deliverable Two functional playground pages. API keys via env. Streaming works.
Week03

Perf, animation, and the polish pass

Core Web Vitals · Framer Motion · Accessibility
DIFFERENTIATOR CWV Motion
Learning Track
  • web.dev Core Web Vitals course (LCP, INP, CLS) INP replaced FID in 2024 — know the difference
  • Lighthouse CI + Vercel Speed Insights — run on your project
  • Framer Motion / Motion docs: layout, variants, AnimatePresence
  • Read: Josh W Comeau on animation easing + physics-based motion
  • next/image, next/font, next/dynamic — when + why
GFE Track
  • Start frontend system design module
  • Deep study: "Design Autocomplete" + "Design News Feed"
  • Adapt both frameworks to "Design an AI playground"
  • Weekend: write your own SD doc for the flagship project
Build Track
  • Add animations: page transitions, card enters, loading states
  • Measure + optimize: LCP < 1.2s, INP < 200ms, CLS < 0.1
  • Lighthouse target: 95+ on all four metrics
  • Add a11y pass: keyboard nav, focus states, ARIA labels, contrast
→ End-of-Week Deliverable Polished playground. Lighthouse ≥ 95. One "wow" interaction.
Week04

Testing, write-up, apply

Vitest · Playwright · Case Study · Application
FINISH LINE Testing Apply
Learning Track
  • Vitest basics: unit tests for utils + hooks
  • Playwright basics: one E2E flow (search → results → click)
  • Write a short case-study blog post on the playground architecture, tradeoffs, perf wins — this becomes your X/LI post
GFE Track
  • 2 full timed mocks: 1 coding + 1 system design
  • Record both, watch back, note filler words + unclear explanations
  • Revisit weak quiz areas identified across weeks 1–3
Build Track
  • Add Vitest tests for 2 key utils/hooks (not exhaustive — signal)
  • Add 1 Playwright E2E covering the happy path
  • Deploy final version. Custom domain. Open-source on GitHub.
  • Write README with architecture diagram + tradeoffs
→ End-of-Week Deliverable Application submitted. Project live. Case study published. Two GFE mocks done.

Build the interview. Not around it.

The fastest path into Mixedbread is to ship something that looks like it could already be in their product. Not a clone — an adjacent tool that demonstrates you understand embeddings, reranking, streaming, and can design a UI around them. This becomes your answer to every application question and your warm intro to the hiring manager.

Project: "Vectorscope" — A Visual Playground for Embeddings & Reranking

An open-source playground that lets you feed text, see how different embedding models project it into vector space, run rerankers across result sets, and stream everything back with motion-rich UI. Positions you as someone who doesn't just consume AI APIs but designs thoughtful interfaces around them.

Must-have features

  • Embedding Visualizer — paste text, see 2D projection (PCA or UMAP via WASM), hover to see raw vectors. Uses Mixedbread's embed API.
  • Reranker Workbench — input a query + 10 candidate docs, watch the reranker sort them in real-time with animated row reorders (Framer Motion layout).
  • Similarity Playground — drag two text chunks into a diff view, see cosine similarity score update live as you edit.
  • Streaming Everything — SSE for all API calls, with skeleton + progressive reveal. No jumpy layouts.
  • Performance — Lighthouse 95+, LCP <1.2s. Instrumented with Speed Insights.

Nice-to-have (time-permitting)

  • Model Comparison — run the same query through 2–3 different embedding models, compare ranking deltas side-by-side.
  • Shareable Sessions — URL state encoding so you can share a specific query+result config.
  • Dark/Light themes — themed with CSS variables, demoing design-system thinking.
  • Keyboard-first — full keyboard nav, cmd+k palette. Signals taste.
→ Stack framework: Next.js 15 (App Router) + React 19
language: TypeScript (strict mode)
styling: Tailwind CSS + shadcn/ui
motion: Framer Motion (Motion library)
monorepo: Turborepo + pnpm workspaces
testing: Vitest + Playwright
ai: Mixedbread API (embed + rerank)
deploy: Vercel + custom domain
analytics: Vercel Speed Insights + PostHog (optional)

What a Mixedbread-style loop looks like — and how to prep each round.

Remote-first AI infra startups run tight loops: 4 rounds, ~4 hours total. Each one tests a different axis. Prep differently for each.

R130 min

Recruiter / Hiring Manager

Background, motivation, why Mixedbread specifically. They're filtering for coherence and genuine interest.

→ Prep
Know their product line. Have a 90-sec pitch. Lead with the playground project.
R260 min

Live Coding (UI Component)

Build something real in 45 min — autocomplete, modal, infinite scroll, data table. JS or TS, React, vanilla browser APIs.

→ Prep
GFE coding medium module. Think out loud. Handle edge cases before asked.
R360 min

Frontend System Design

"Design the playground page for our API" or "Design a streaming chat interface." Whiteboard/Miro-style.

→ Prep
GFE SD module. Have a framework (requirements → architecture → components → data → optimize).
R445–60 min

Culture + Deep Dive

Walk through your flagship project. Hard technical follow-ups. Values fit. "What would you change about our site?"

→ Prep
Audit mixedbread.com yourself. Have 3 specific, kind critiques ready.

The form has two open-ended questions. They matter more than the resume.

Most candidates fill these out as afterthoughts. Treat them as mini-essays. Specificity beats eloquence.

Proud Project
Lead with Vectorscope (the flagship you build in §04). Frame it around one concrete tradeoff — "I rebuilt the reranker viz from state-driven animations to layout animations to hit 60fps on long lists." Specificity signals engineering depth. Link the live demo, the repo, and the case-study post. End with one line on what you'd do next if given more time.
Why Mixedbread
Don't list their product names back at them. Pick one specific thing — e.g., "Your reranker + embedding positioning is the rare AI infra play where the frontend actually matters, because developers evaluate through the playground" — and build the answer around it. Mention someone on the team you've seen post online (if genuine). End with: "I built Vectorscope specifically because I wanted to feel what building on your API is like before applying. Here's what I learned."
LinkedIn / X / GitHub
Post the case study on both LinkedIn and X the day before you submit. Tag Mixedbread. This means when they check your profile, the top post is your project. Pin the GitHub repo. If your X bio still says "founder, Lookover" — fine, but add "building on @mixedbread this week" for 2 weeks. Signal fit before the form submits.

The cheap mistakes that cost the offer.

→ Do
  • Ship the playground live on a custom domain before applying.
  • Record a 60-sec Loom walkthrough. Link it in the application.
  • Use Mixedbread's actual API, not OpenAI's — proof of product-fit.
  • Write one case-study post about the build. Link it.
  • Run Lighthouse publicly. Screenshot the scores. Paste them in the post.
  • Audit mixedbread.com and keep 3 polite critiques in your back pocket for R4.
→ Don't
  • Don't submit a generic resume. Tailor it to mention playground, perf, AI.
  • Don't skip the a11y pass. They ship to devs — devs notice.
  • Don't use `pages/` router in the flagship project. They said "App Router preferred."
  • Don't fake AI expertise. Saying "I understand what embeddings are for" beats jargon-stuffing.
  • Don't over-engineer. 4 polished features > 12 half-built ones.
  • Don't forget the Lookover + agency context — frame it as founder-mode, not distraction.