Expertise — 02
Job Hunter Automation
Asynchronous pipeline for semantic discovery and ATS automation
The Problem
The modern job market is saturated with high-noise aggregators, making manual discovery and application to quality roles a massive operational bottleneck.
Architecture
Orchestration
Prefect
6-stage asynchronous workflow
Scraping
Playwright (Async)
Career page detection & stealth crawling
Evaluation
all-MiniLM-L6-v2
Local semantic similarity scoring
Generation
WeasyPrint / Jinja2
Tailored PDF resume construction
Integration
Telegram API
Actionable notifications & callbacks
Implementation
Implemented a multi-node pipeline using asyncio and ThreadPoolExecutor to handle concurrent domain crawling and parallel job evaluation. The system uses local Sentence-Transformers for privacy-preserving semantic matching, bypassing the need for cloud LLMs. High-scoring jobs trigger a dynamic generation engine that selects tech-stack specific bullets from a master resume and builds a tailored PDF, which is then served via Telegram with interactive 'Apply' triggers for ATS platforms.
Manual Work
-90%
Time Saved
Privacy
100%
Local ML
Coverage
100+
Domains/Run
Key Learnings
"Local execution of semantic models proved that complex decision-making could be achieved without external API overhead, maintaining 100% data privacy while drastically reducing system latency."