AXIS - Agentic Resume Workspace

AXIS is an AI agent workspace for resume generation and refinement. It turns raw career materials, uploaded resumes, and target job descriptions into structured JSON Resume content, then routes the work through generation, JD matching, field-level rewriting, compression, LaTeX rendering, PDF export, cloud snapshots, and public sharing.

Why AXIS

• Agentic workflow instead of one fixed prompt.
• Schema-first resume data instead of fragile free text.
• RAG memory over projects, skills, previous versions, and target roles.
• Field-level editing and version rollback before PDF export.
• Production-oriented Flask APIs with auth, usage limits, sharing, subscriptions, and payments.

Agent Architecture

AXIS separates intent, memory, task execution, and rendering.

flowchart LR
  U[User input / files / JD] --> API[Flask API]
  API --> RAG[RAG memory retrieval]
  API --> WF[LangGraph Resume Workflow]
  RAG --> WF
  WF --> IR[Intent Routing]
  IR --> G[Full generation]
  IR --> M[JD matching]
  IR --> L[Local rewrite]
  IR --> C[Compression]
  G --> JSON[JSON Resume]
  M --> JSON
  L --> JSON
  C --> JSON
  JSON --> PDF[LaTeX / PDF Renderer]
  JSON --> DB[(Snapshots + Vector Memory)]
  PDF --> FE[React editor preview]

Loading

Core backend pieces:

Area	File	Role
Agent workflow	backend/services/resume_agent_workflow.py	LangGraph task orchestration for import, generation, continuation, local edits, retargeting, compression, and polishing
Chat agent	backend/services/resume_chat_agent.py	Chat-panel intent handling, trace-friendly state, and response writing
RAG memory	backend/services/resume_memory_service.py	Embedding, vector storage, retrieval, and context formatting
Resume pipeline	backend/services/resume_generation_pipeline.py	Stable service boundary used by routes and graph nodes
Renderer	backend/services/latex_service.py	Structured JSON Resume to LaTeX/PDF

RAG Memory

Saved and generated resumes are indexed into resume_memory_vectors:

• project and work experience chunks
• skill keywords
• awards and optimization summaries
• target JD and role instructions

The default embedding is deterministic and database-backed, so local tests and low-cost deployments work without another vector service. The storage boundary can later be swapped for pgvector or an external vector database without changing the agent workflow contract.

Product Surface

Module	Capability
Editor	Agent chat, file input, field editing, template/layout controls, PDF preview
Resume tools	parsing, structured generation, JD matching, local rewrite, compression, LaTeX compile, PDF export
Persistence	cloud save, snapshots, rollback, public share links
Account	email verification, login, OAuth hooks, password reset, profile management
Billing	Free/Pro/Max entitlements, Stripe, Alipay, Bitcoin unique-amount flow

Tech Stack

Layer	Stack
Frontend	React 18, Vite, React Router, i18next, TipTap, react-pdf
Backend	Flask 3, LangGraph, SQLAlchemy 2, Alembic, Pydantic, PyJWT
AI	OpenAI-compatible chat API, Google AI Studio fallback
Data	Postgres in Docker, SQLite fallback, DB-backed vector memory
Render	LaTeX template renderer and PDF compilation
Deploy	Docker Compose, Gunicorn backend, Nginx frontend

Architecture Docs

Getting Started

cp .env.example .env
docker compose up --build

Open:

• Frontend: http://localhost:3000
• Backend: http://localhost:5001

Run locally without Docker:

python -m venv .venv
source .venv/bin/activate
pip install -r backend/requirements.txt
alembic upgrade head
python -m flask --app backend.app run --host 0.0.0.0 --port 5001 --debug

cd frontend
npm install
npm run dev

Production-style compose:

docker compose -f docker-compose.prod.yml up --build

Checks

python -m pytest backend/tests
cd frontend && npm run build