Why Keyword-Based ATS Is Dead: Engineering Autonomous AI Evaluation in NiteHire

### The Flaw of Traditional Resume Screening Candidates have learned to beat legacy ATS software by stuffing white-font keywords or prompting ChatGPT to rewrite their resumes to match every job description bullet. The result is recruiters receiving 500 'perfect' resumes, 90% of which cannot pass a basic architectural whiteboard interview. ### The NiteHire 2-Level Screening Engine In building **NiteHire**, I separated candidate evaluation into two non-overlapping, mathematically grounded tiers: 1. **Level-1 Semantic Ingestion (Groq Llama 3.3 70B)**: Real-time parsing of project complexity, timeline coherence, and technical depth rather than raw keyword frequencies. Latency stays under 450ms. 2. **Level-2 Interactive Dynamic Screening (Gemini 2.5 Multi-Modal)**: Interactive candidate interview where the AI adapts questions based on the candidate's previous responses, actively checking for genuine experiential depth. This ensures companies hire true technical builders rather than prompt-engineering resume spammers.