TalentShift

The Problem

HR teams manually review hundreds of resumes per role, copying candidate details into spreadsheets for comparison. The process is slow, inconsistent, and prone to overlooking qualified candidates buried in non-standard resume formats.

The Solution

I built TalentShift as a resume parsing and evaluation platform.

• Resume upload and parsing — Accepts PDF/DOCX resumes, extracts text, and sends it to OpenAI for structured data extraction (name, contact, experience, skills, education).
• Structured profiles — Parsed data is stored in Supabase as searchable, filterable candidate profiles.
• AI-powered dashboard — HR users can filter candidates by skill, experience level, education, and get AI-generated summary comparisons between shortlisted candidates.
• Batch processing — Upload multiple resumes at once for bulk ingestion.

What Went Wrong

Early on, the OpenAI extraction was inconsistent across resume formats. Creative resumes with non-standard layouts (two-column, infographic-style) would produce partially parsed or incorrectly attributed data — skills listed under education, job titles mixed with company names.

The fix: I added a two-stage extraction pipeline. Stage 1 normalizes the raw text into a consistent format using a dedicated prompt. Stage 2 extracts structured fields from the normalized text. This separation dramatically improved accuracy across diverse resume formats.

Results

• Automated resume-to-profile conversion in seconds
• Searchable and filterable candidate database
• Handles diverse resume formats via two-stage extraction