桦 · Personal AI Workspace

An extensible personal command center for projects, tasks, knowledge, documents, automations, AI workflows, and career development. The original AI Resume Generator and Job Application Tracker are preserved as the Career workspace.

The new product information architecture, shared data contracts, and staged Supabase migration are documented in docs/personal-ai-workspace-architecture.md.

Cloud deployment instructions are in docs/cloud-deployment.md.

Features

Smart JD Parsing — Extracts skills, keywords, responsibilities, soft requirements, deal-breakers from any job description. Includes noise cleaning to remove benefits/legal/salary sections and save API tokens.
Intelligent Bullet Retrieval — 5-priority strategy (deal-breaker match, responsibility mirror, keyword density, soft requirement proof, bonus signal match) to select the most relevant bullets from your profile.
LaTeX Resume Generation — Fills your LaTeX template with tailored content, mirrors JD terminology and action verbs, bolds key metrics.
ATS Scoring + Auto-Optimization — Keyword matching + Claude semantic analysis. Iteratively refines the resume (up to 3 rounds) until it meets score thresholds (overall: 80, keyword: 60%, relevance: 80, impact: 80).
Cover Letter Generation — Produces a tailored 3-4 paragraph cover letter that addresses the JD's technical and soft requirements.
Multiple Templates — Switch between classic, modern (blue accents), and consulting (conservative) styles.
Job Search — Search Indeed and Adzuna, deduplicate results, and rank jobs against your profile with Claude.
Application History — Tracks every resume generated: company, role, ATS scores, template, and application status. Expandable detail view with content preview.
PDF Compilation — Compiles LaTeX resumes and cover letters to PDF via pdflatex, downloadable from both CLI and web UI.
Web UI — Full-featured Next.js frontend with FastAPI backend: dashboard with stats, generate page with step-by-step progress, job search, template picker, profile management, and file downloads (LaTeX/PDF/TXT).

Architecture

profile.json + JD text input
              |
    [1] JD Noise Cleaner    -> removes benefits, legal, salary noise
    [2] JD Parser            -> extracts structured requirements (Claude API)
    [3] Bullet Retriever     -> selects top-k relevant bullets (Claude API)
    [4] Resume Generator     -> fills LaTeX template (Claude API)
    [5] ATS Scorer           -> keyword + semantic scoring (Claude API)
         |--- below threshold? -> refine and re-score (up to 3 rounds)
    [6] Cover Letter         -> tailored cover letter (Claude API)
    [7] PDF Compiler         -> pdflatex compilation
              |
    output/*.tex, *.txt, *.pdf, *_cover_letter.txt

Quick Start (Web UI)

# Windows: one-click launcher
start.bat

# Or manually:
# Terminal 1 — Backend
pip install -r requirements.txt
uvicorn api.server:app --reload --port 8000

# Terminal 2 — Frontend
cd frontend && npm install && npm run dev

Open http://localhost:3000 in your browser.

How I Used Codex to Build This Project

I used Codex as a coding agent to move Auto-Resume from a local resume prototype toward a deployable personal AI workspace. This is separate from the product's runtime AI: the application uses the Anthropic API to analyze jobs and generate career materials.

The repository history records this work in agent/* and codex/* branches. Between July 15 and July 21, 2026, I opened 11 agent-assisted pull requests; PRs #1–#9 were merged through GitHub, while #10–#11 remain open.

Pull requests	Recorded Codex-assisted work
#1	Moved a machine-specific Notion/Indeed job tracker into the repository, replaced local paths with portable ones, documented setup, and protected credentials and runtime files.
#2	Completed the secure MVP across 7 commits: cloud workspace APIs, career automation, OAuth, invite-only registration, persistent quotas, durable generation jobs, safer PDF compilation, editable document history, UI refinement, CI, and regression tests.
#3–#4	Diagnosed two Vercel deployment problems from their root causes: excluded frontend source files and stale client API URLs bypassing the same-origin proxy.
#5–#7	Built user-facing career workflows: full profile editing, restored Indeed + Adzuna search with source labels and graceful fallback, and explicit multi-job selection before spending AI quota on resume generation.
#8–#9	Hardened AI and deployment reliability by retrying truncated structured responses and ensuring the three LaTeX templates are included in the Vercel bundle.
#10–#11	Proposed a narrowly allow-listed PDF glyph-map fix, separated job-match scores from generated-resume ATS scores, and improved the first-run career experience.

The PR record also shows how I worked with Codex:

Each PR starts from a concrete product problem or production failure and states the intended outcome.
Several PR descriptions record a root-cause investigation before the fix, including the missing Vercel source bundle, proxy bypass, absent Indeed integration, missing user-selection step, and excluded LaTeX templates.
Changes were made on isolated branches, including backend, frontend, tests, and documentation when the task crossed those boundaries.
Each PR records its relevant verification. Across these PRs, the backend suite grew from 16 to 23 tests; validation also included ESLint, Hua UI rules, Next.js production builds, Python compilation, live JobSpy checks, PDF generation, and diff checks.
Accepted work reached main through GitHub PR merges; the two unmerged changes remain isolated on their branches.

This history reflects how I use Codex: not just to generate code, but to investigate failures, preserve safety boundaries, add regression coverage, and leave each change in a reviewable state. I remain responsible for the requirements, product decisions, credentials, deployment approval, and final merge.

Setup

# 1. Install dependencies
pip install -r requirements.txt
cd frontend && npm install && cd ..

# 2. Create .env with your API key
echo "ANTHROPIC_API_KEY=sk-ant-..." > .env

# 3. (Optional) For job search, add Adzuna keys — free at https://developer.adzuna.com/
echo "ADZUNA_APP_ID=your_id" >> .env
echo "ADZUNA_APP_KEY=your_key" >> .env

# 4. (Optional) Install pdflatex for PDF output
# Windows: install MiKTeX — https://miktex.org/download

Web UI

The web frontend provides a complete interface for all features:

Dashboard — Stats, application history with expandable rows to preview and download generated files
Generate — 3-step flow: paste JD, analyze, generate & optimize. Download resume PDF, LaTeX, cover letter
Job Search — Search Adzuna jobs, Claude-ranked by profile fit with match scores
Templates — Visual template picker (classic, modern, consulting) — click to generate
Profile — Manage personal info, skills, experience, and projects

Dashboard Detail View

Click any history row to expand and access:

Resume PDF / LaTeX source download
Cover Letter PDF / TXT download
Tab preview of resume LaTeX and cover letter content
Inline status updates (generated → applied → interview → offer/rejected)

CLI Usage

Generate Resume + Cover Letter

# From a JD file
python main.py generate --jd path/to/jd.txt

# Interactive mode (paste JD in terminal)
python main.py generate

# Choose a template
python main.py generate --jd jd.txt --template modern

# Skip cover letter
python main.py generate --jd jd.txt --no-cover-letter

# More bullets
python main.py generate --jd jd.txt --top-k 15

Search Jobs

python main.py search --query "Machine Learning Engineer" --location canada
python main.py search --query "Data Scientist" --location us --top-n 5

View Application History

python main.py history
python main.py history --update 1:applied
python main.py history --update 1:interview

Status flow: generated -> applied -> interview -> offer / rejected

List Templates

python main.py templates

Available: classic (default), modern (blue accents), consulting (conservative)

Output

Each run produces:

output/
  20260309_143022_ml_engineer.tex             # LaTeX source
  20260309_143022_ml_engineer.txt             # Plain text (for Overleaf)
  20260309_143022_ml_engineer.pdf             # Compiled PDF
  20260309_143022_ml_engineer_cover_letter.txt # Cover letter

API Endpoints

Method	Path	Description
GET	`/api/health`	Health check
GET	`/api/profile`	Get profile with stats
PUT	`/api/profile/personal`	Update personal info
PUT	`/api/profile/skills`	Update skills
POST	`/api/profile/experience`	Add experience
DELETE	`/api/profile/experience/{id}`	Delete experience
POST	`/api/profile/project`	Add project
DELETE	`/api/profile/project/{id}`	Delete project
GET	`/api/templates`	List templates
POST	`/api/parse-jd`	Parse job description
POST	`/api/retrieve-bullets`	Select relevant bullets
POST	`/api/generate`	Generate resume + cover letter
POST	`/api/score`	ATS score a resume
POST	`/api/refine`	Refine resume with ATS feedback
POST	`/api/generate-full`	Full pipeline in one call
POST	`/api/generation-jobs`	Queue an idempotent generation
GET	`/api/generation-jobs/{id}`	Poll generation status/result
GET	`/api/profile/completeness`	Validate profile readiness
GET	`/api/history`	List history (light)
GET	`/api/history/{id}`	Get record with full content
PATCH	`/api/history/{id}`	Update status
POST	`/api/history`	Save new record
POST	`/api/compile-pdf`	Compile LaTeX to PDF
POST	`/api/compile-cover-letter-pdf`	Compile cover letter to PDF
POST	`/api/search-jobs`	Search + rank jobs

Customizing Your Profile

Edit data/profile.json. Each bullet needs:

{
  "id": "b001",
  "text": "Built a production RAG system processing 100+ meetings/week...",
  "tags": ["RAG", "NLP", "Python"]
}

Tags help the retriever match bullets to JD keywords more accurately.

Project Structure

main.py                     # CLI entry point with subcommands
api/
  server.py                 # FastAPI backend (REST API)
frontend/
  src/app/                  # Next.js App Router pages
    page.tsx                # Personal workspace dashboard
    career/                 # Career overview, applications, and interview prep
    projects/page.tsx       # Project workspace
    tasks/page.tsx          # Task workspace
    knowledge/page.tsx      # Knowledge workspace
    documents/page.tsx      # Generated documents and versions
    automations/page.tsx    # Schedules and run history
    integrations/page.tsx   # Connected services
    generate/page.tsx       # Generate resume flow
    search/page.tsx         # Job search
    templates/page.tsx      # Template picker
    profile/page.tsx        # Profile management
  src/components/           # Shared components (Sidebar, Header)
  src/lib/api.ts            # API client
src/
  jd_parser.py              # JD noise cleaning + structured parsing
  retriever.py              # 5-priority bullet selection
  generator.py              # LaTeX resume generation + refinement
  ats_scorer.py             # Keyword + semantic ATS scoring
  cover_letter.py           # Cover letter generation
  templates.py              # Multi-template manager
  job_finder.py             # Indeed and Adzuna job search
  history.py                # Application history tracking
data/
  template.tex              # Classic LaTeX template
  template_modern.tex       # Modern template (blue accents)
  template_consulting.tex   # Consulting template (conservative)
  profile.json              # Your profile (gitignored)
  sample_jd.txt             # Example JD
start.bat                   # One-click launcher (Windows)

Tech Stack

AI: Claude API (claude-sonnet-4-6 by default, configurable with ANTHROPIC_MODEL)
Backend: FastAPI + Uvicorn
Frontend: Next.js 16 (App Router) + TypeScript + Tailwind CSS
Resume: LaTeX templates + pdflatex compilation
Job Search: Indeed via JobSpy, with optional Adzuna API results
Icons: Custom Birch icon set

Roadmap

Scheduled career discovery, deduplication, review items, material generation, retries, notifications, and hosted worker setup are documented in Career automation workflow.

Name		Name	Last commit message	Last commit date
.github/workflows		.github/workflows
.vscode		.vscode
api		api
automation		automation
browser-extension		browser-extension
data		data
deploy		deploy
docs		docs
frontend		frontend
integrations		integrations
openclaw		openclaw
scripts		scripts
src		src
tests		tests
.dockerignore		.dockerignore
.env.example		.env.example
.gitignore		.gitignore
.vercelignore		.vercelignore
Dockerfile		Dockerfile
Procfile		Procfile
README.md		README.md
main.py		main.py
railway.json		railway.json
requirements.txt		requirements.txt
server.py		server.py
start.bat		start.bat
vercel.json		vercel.json

Navigation Menu

Repository files navigation

桦 · Personal AI Workspace

Features

Architecture

Quick Start (Web UI)

How I Used Codex to Build This Project

Setup

Web UI

Dashboard Detail View

CLI Usage

Generate Resume + Cover Letter

Search Jobs

View Application History

List Templates

Output

API Endpoints

Customizing Your Profile

Project Structure

Tech Stack

Roadmap

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

桦 · Personal AI Workspace

Features

Architecture

Quick Start (Web UI)

How I Used Codex to Build This Project

Setup

Web UI

Dashboard Detail View

CLI Usage

Generate Resume + Cover Letter

Search Jobs

View Application History

List Templates

Output

API Endpoints

Customizing Your Profile

Project Structure

Tech Stack

Roadmap

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages