AI / NLP

CV Transformer

An AI system that turns casual notes about your experience into polished, professional CV entries - running locally on a Polish language model.

Year2026
RoleAI Developer
DurationMay–June 2026
PythonHugging Face TransformersGradioPyTorchLLMNLP
CV Transformer
CV Transformer — overview
Overview

The story behind the project

The challenge
Problem

Plenty of people have solid experience but can't put it into "official" CV language - a note like "did bookkeeping, checked invoices, lived in Excel" gets lost instead of becoming a strong CV line.

The approach
Solution

I built a pipeline around Bielik-1.5B-v3.0-Instruct, a Polish instruction-tuned LLM from SpeakLeash, running locally on GPU. Notes go through a one-shot prompt, and the output passes through post-processing that strips markdown artifacts. I also added a second feature: upload a full CV file (PDF/DOCX/TXT), auto-split it into sections, and rewrite every entry. Both live in a two-tab Gradio interface.

0.696
The result
Outcome

Average semantic similarity (cosine similarity) of 0.696 and a compression ratio of 0.89 - the model tightens text just enough for CV style without losing meaning. Tested four temperature values to balance creativity against consistency.

Features

What CV Transformer does

Feature 01

Real-Time Note Transformation

Type a casual note about something you did at work and the model rewrites it into a polished, CV-ready sentence in seconds - powered by a carefully engineered one-shot prompt rather than fine-tuning.

  • Real-time transformation of a single note into a professional CV entry
  • One-shot prompting to eliminate hallucinations
  • Post-processing that cleans formatting artifacts
Real-Time Note Transformation
Feature 02

Full CV Upload & Rewrite

Upload an entire CV as PDF, DOCX, or TXT and the pipeline automatically splits it into sections and rewrites every entry in one pass - no copy-pasting line by line.

  • Upload a CV file and auto-improve every section at once
  • Automatic section splitting for PDF/DOCX/TXT files
  • Interactive two-tab Gradio interface
Full CV Upload & Rewrite
Tech Stack

Built with

PythonHugging Face TransformersGradioPyTorchLLMNLP