CV Enhancer
AI-powered tool for Indonesian job seekers to optimize their resumes for job descriptions and ATS screening without fabricating facts.
Project Overview
CV Enhancer is the flagship tool in the wizdev.tech ecosystem. It is built to bridge the gap between job seekers’ actual qualifications in Indonesia and the way their experience is presented in application materials.
Many applicants possess valuable experience but get filtered out early by ATS parsers or initial recruiter screening due to vague wording, misaligned keywords, or unreadable document formatting.
Fundamental principle: AI is strictly forbidden from hallucinating experiences, inventing metrics, or adding fictitious skills. Optimization focuses entirely on clarity of language, achievement structuring, and keyword alignment against target job descriptions.
Problems
- Low Job Description Alignment: Applicants frequently submit a single generic resume to dozens of distinct openings.
- Weak Impact Presentation: Real achievements are often described merely as routine task lists without demonstrating methodology or results.
- Generic AI Hallucinations: Prompting public chatbots often introduces fabricated tools or exaggerated claims, compromising the candidate during interviews.
- ATS Unfriendly Formatting: Complex multi-column layouts cause text extraction failures in company parser software.
Target Users
- Fresh Graduates and Vocational Alumni: Entry-level seekers translating academic and internship projects into professional terminology.
- Career Switchers: Workers transitioning fields who need to highlight transferable skills effectively.
- Active Skilled Job Seekers: Candidates who need rapid resume tailoring across multiple targeted vacancies.
Technical Decisions
- Serverless on Cloudflare: Migrated to Cloudflare Workers (Hono API) and Cloudflare Pages (React + Vite) for low latency across Indonesia and low operating costs.
- Distributed Storage: Using Cloudflare D1 for edge SQL and Cloudflare R2 for temporary file staging with automated cleanup immediately after text extraction.
- Strict Prompt Engineering: Inference pipelines (Claude API) enforce strict guardrails: fact extraction exclusively from the user’s CV and automatic refusal to invent claims.
- Focused Minimalist Interface: Clean layout with step-by-step navigation without distracting elements.
Challenges
- Data Integrity: Ensuring the model never fabricates employment dates, positions, or unmentioned skills.
- Indonesian Document Parsing: Extracting clean text from non-standard PDF and DOCX files.
- User Privacy: Minimizing collected data and ensuring no resume data is used for public model training.
Measurable Results
- Zero-Hallucination Pipeline: Verification rules ensure every generated bullet maps to original source facts.
- Fast Processing: End-to-end document extraction and structured enhancement completes in seconds on edge workers.
- Privacy Enforcement: Raw uploaded documents are purged immediately after processing.