ApplAI — AI-Powered Talent Intelligence Platform
Overview
ApplAI is an end-to-end talent intelligence platform designed to bridge the gap between job candidates and hiring teams. It automates unstructured resume ingestion, parses multi-language candidate backgrounds into verified skills and experiences, and performs bidirectional ranking between candidate profiles and job requisitions. The project was presented as an academic research poster at University of San Andrés.
Key Features
- Custom Fine-Tuned LLM (Qwen Family): Adapted and fine-tuned a specialized language model to parse diverse PDF/Word CV layouts and extract canonical entity schemas without hallucinations.
- Hybrid Search Engine: Combines dense vector embeddings with lexical BM25 keyword matching to retrieve candidates with both exact keyword prerequisites and semantic background alignment.
- Fallback Ranking Strategy: High-availability resilience layer that gracefully switches to lightweight heuristic and cached vector rankings when upstream LLM APIs hit rate limits.
- Production-Grade Fullstack Architecture: Built with a Next.js frontend, asynchronous FastAPI microservices backend, and Supabase PostgreSQL with pgvector for vector persistence.
Architecture & Implementation
User CVs are uploaded through Next.js to secure cloud storage. A FastAPI worker service extracts text, normalizes tokens, and invokes the fine-tuned Qwen model on Google Cloud to extract structured JSON (skills, seniority, education, roles). Vectors are indexed into Supabase (PostgreSQL + pgvector). A bidirectional matching service queries the hybrid search index and returns ranked candidate lists with explainable match breakdowns.
Results & Performance
Rigorous evaluation across a test benchmark of 100 diverse CVs and 10 real-world job specifications demonstrated a 95% accuracy rate in candidate-job category assignment, drastically cutting manual screening effort while preserving fair matching standards.