~/projects/resume-builder

Resume Builder

> A local-first AI resume builder that turns structured experience data into tailored, print-ready LaTeX resumes — powered by an on-device LLM.

view_source

// pipeline

resume.db (SQLite)structured source of truth
+ job_description.txttarget role context
→ llama.cpp (local)tailor + rephrase bullets
→ resume.textemplated LaTeX
→ resume.pdfnative export
Type
Dev Tool
Role
Solo build
Frontend
React + Tauri
Backend
Rust
Storage
SQLite + sqlite-vec
Inference
llama.cpp (QLoRA 3B)
Status
In progress
RustTauriReactTypeScriptSQLitellama.cppQLoRALaTeX

Local-first LLM

Runs entirely offline against an embedded llama.cpp runtime, with semantic search over your content via ONNX embeddings in sqlite-vec — no API keys, no data leaving the machine.

Structured → LaTeX

Stores resume content in SQLite, then compiles a role-tailored LaTeX resume to PDF on-device. Consistent typography, zero manual formatting.

Live editor

Edit bullet points and have the model rephrase, tighten, or quantify them inline. Diff-style preview before anything is committed to the document.

Distilled local model

Distilled Claude outputs into a QLoRA fine-tuned Qwen2.5-3B, quantized to a ~1.8 GB GGUF that writes a grounded document in ~30 s on CPU — halving LLM-judged fabricated claims (98 → 46) vs. the base model on 20 held-out job postings.

Cross-platform CI

CI builds 5 installer formats across Windows, macOS, and Linux, backed by Rust unit and golden-file tests.

Native PDF export

Tauri-bundled Rust backend shells out to a LaTeX toolchain and returns a print-ready PDF — a single binary, no Electron bloat.