A multilingual career exploration site that converts vague job names into real-person growth paths, ordered preparation steps, and honest reality. Content production runs on an AI pipeline.

Positioning

Reputo is a multilingual career exploration platform that turns vague job aspiration into a comprehensible preparation path for students and early career explorers. Not an encyclopedia, not a counseling service — an exploration site where role-model narratives, step-by-step roadmaps, and reality checks live in a single flow.

Market and Problem

Students can name jobs they admire but rarely understand what those roles require or where preparation begins. Existing career content tends to be (1) abstract like a job dictionary, (2) success-story-heavy without reality, or (3) gated behind expensive admissions consulting.

Reputo reduces that uncertainty by linking role-model narrative + execution roadmap + reality check within a single page. See “the person you want to become,” follow “the stages of their growth,” and decide “the next action you can actually start” — all in one flow.

Core Audience and Personas

Middle / high school and university students — Users who want both inspiration and actionable signals beyond vagueness
Early career explorers — People interested in a role but unsure of preparation order
Parents, teachers, mentors — Adults who need clear, structured material to discuss with students
Multilingual student audiences — ko / en / ja / zh-cn / zh-tw, each version emphasizing region-relevant companies and figures

Value Proposition and Differentiation

Role model + roadmap + reality on one page — Other career sites split person stories / job descriptions / preparation guides into separate pages. Reputo connects “how this person reached the role → what stages along that path look like → what challenges to expect realistically” on a single page
Five languages, fully equivalent — Korean / English / Japanese / Simplified Chinese / Traditional Chinese. Each version is not just a translation — region-specific companies and figures are emphasized (e.g., Korean = Samsung / Naver / Kakao, Japanese = Sony / Toyota, zh-cn = Alibaba / Tencent)
AI pipeline for content scale — researcher.py (research) → writer.py (drafting) → editor.py (review / polish) → regenerate_people.py (role model refresh) — a four-stage automation that lets a single operator maintain a 5 languages × N roles × M people content matrix
Free, no-friction entry — Browsing requires no signup; lightweight authentication only for interactive features (quiz / mentoring)

Core User Flow

Field exploration: Start at a category — Technology / Business / Healthcare / Finance / Design — and narrow
Role model discovery: Pages for real people in roles of interest (growth story, decisive moments, failure and recovery)
Roadmap study: Sequence of preparation → learning → experience accumulation, plus recommended materials
Reality check: Rewards, challenges, common misconceptions, what year 1–3 actually looks like
Decide next step: Career quiz to narrow interest / recommended reading / 1:1 mentoring (live professional connection)

User flow — from aspiration to action

flowchart LR
    Browse["Field exploration
(Tech · Business · ...)"]
    Person["Role model discovery
growth story · decisive moments"]
    Road["Step-by-step roadmap
prep → learn → experience"]
    Real["Reality check
rewards · challenges · misconceptions"]
    Next["Next step
quiz · books · mentoring"]

    Browse --> Person
    Person --> Road
    Road --> Real
    Real --> Next
    Next -.->|narrowed interest| Browse

A closed loop of explore → validate → act → re-explore. A single visit doesn’t end the relationship — the product accompanies the entire decision process.

Business Model Hypothesis

Pricing is being validated in stages.

1:1 mentoring matching — Connect students directly with working professionals on an hourly basis. Free content builds conviction; mentoring carries the action-stage fee
School / institution licensing — Schools curate Reputo content for their students under their own white-label or integration
Career book / program affiliate — Recommended books, bootcamps, and certification programs as partner placements

System Architecture (planning decisions become system structure)

Two planning decisions drove the architecture.

1. “A 5-language content matrix must be sustainable by a single operator.” → Direct authoring is impossible at this scale, so a four-stage AI pipeline carries the load: researcher gathers material, writer drafts, editor polishes, regenerate_people refreshes person pages. The operator handles only inputs (job / person candidates) and review of outputs. The only path to keeping all five languages at equivalent quality.

2. “An exploration site should be static for SEO and speed.” → Hugo static build + Tailwind. Content lives in markdown + JSON, and a single build generates the entire 5-language site. pagefind handles search via a client-side index — no search server required for 5-language search.

flowchart TD
    Source["Input candidates
(roles · people · fields)"]
    Pipe1["researcher.py
research + verification"]
    Pipe2["writer.py
drafting
(5 languages)"]
    Pipe3["editor.py
review · polish
(equivalent across 5 langs)"]
    Pipe4["regenerate_people.py
person page refresh"]
    Content[("content/
markdown + JSON")]
    Hugo["Hugo build
(5 languages together)"]
    PF["pagefind
(client-side search index)"]
    CDN["CloudFront → S3
reputo.net"]

    Source --> Pipe1
    Pipe1 --> Pipe2
    Pipe2 --> Pipe3
    Pipe3 --> Pipe4
    Pipe4 --> Content
    Content --> Hugo
    Hugo --> PF
    Hugo --> CDN
    PF --> CDN

Content production via AI pipeline, site delivery via static build, search via client-side index — operating cost is compressed at every layer.

Technology Choices and Trade-offs

Site: Hugo + Tailwind + PostCSS. Five languages built simultaneously. Different from the workspace’s Vite + React standard — content sites win on SEO, speed, and multilingual operations with static builds
Content pipeline: Python (researcher / writer / editor / regenerate_people / image_downloader / job_recommender) using external LLM APIs. The layer that makes 5 langs × N roles × M people maintainable solo
Search: pagefind (client-side index). Five-language search with no server
Data: Structured data (jobs.json etc.) for role / field mapping; person pages live as content/people/<slug>/index.{ko,en,ja,zh-cn,zh-tw}.md (5 files per person)
Hosting: AWS S3 + CloudFront, workspace standard
What was dropped: User signup (browsing is anonymous; only mentoring requires auth), realtime recommendations / algorithmic feed (preserves static-site advantages), self-hosted LLM (external APIs win on cost and quality)

Operational Automation

Content pipeline auto-runs — Input candidate queue → 4-stage automation → markdown output. Operator handles review only
5-language completeness lint — Every page must have all five language files; missed translations fail the build (the multilingual rule from CLAUDE.md as a hard guardrail)
Image auto-download — image_downloader.py collects person / company / role images and verifies licensing
Site deployment — main push → Hugo build → S3 upload + CloudFront invalidation, automated
Content checks — content_checker.py pre-validates language completeness, broken links, and missing metadata

Current State and Operational Signals

Status: Live. Five languages running at parity, content pipeline stabilized
Started: 2025-09 (content pipeline buildout); launched 2026-01
Infrastructure: AWS S3 + CloudFront (static); Python content pipeline
Verification signals: Per-language traffic ratios (Korea vs global vs Japan vs Chinese-speaking regions), career quiz completion rate, and average dwell time on person pages

Retrospective and Next Hypotheses

What worked: AI pipeline + 5 languages + static build combined into a structure where a single operator can maintain the content matrix — what would take a year of manual writing fits into a week. The “role model + roadmap + reality on one page” structure is validated by student dwell time
What I would redo: Launching all five languages at once stretched review capacity (one operator can’t simultaneously vet five languages). Stabilizing ko / en first, then sequentially adding ja / zh would have been safer
Next hypotheses: (1) 1:1 mentoring matching (live professional connection), (2) School / institution licensing (curated white-label), (3) Career book / bootcamp affiliate, (4) Mobile PWA / app (after usage pattern validation)

Comparable Engagements

The capabilities developed solo on Reputo transfer cleanly to other domains.

AI-pipeline multilingual content sites — Maintaining N × M × L content matrices solo (encyclopedias, guides, comparison sites, catalogs)
Role model / case / roadmap content UX — Translating abstract information into person stories + ordered actions for any domain (career, self-development, learning, health)
Four-stage AI researcher → writer → editor content automation — From a single input to reviewable output, with five-language parity guaranteed
Hugo multilingual static site + pagefind search — Server-less five-language search site operations
JSON data + markdown content separation model — Decoupling structured data (role / field mapping) from narrative content (people, roadmaps) so operations stay simple

I prefer engagements where one person carries the work end to end. Reach me via /work-with-me or /contact.