feat: implement 5 high-priority premium recruiter features with full unit tests and deduplication

Remove Pool Statistics Dashboard section from frontend

fix(pipeline): resolve double-waiting bug on frontend animation setTimeout for pipeline steps

fix(compare): resolve candidate_id and id key mismatch bug in recruitment console compare action

Enhance SwarmMatrix with premium recruiter features (AI interview guides, side-by-side comparative matrix, recruiting brief exporter, fuzzy location filters) and modernise query expansion Gemini SDK

fix(ui): redesign candidate cards layout to resolve spacing and overlapping

Revert secondary name validation and extraction fallback to restore original pipeline

Support GEMINI_SECONDARY_MODEL to decouple secondary name extraction rate limits from main pipeline

Implement filename-based name extraction fallback and backoff jitter to prevent Gemini API 429 rate limits under high concurrency

Add Hugging Face metadata frontmatter configuration to README.md

Enhance Talent Radar with dynamic dates, sector-neutral scoring/parsing, premium UI dashboard, and full unit test coverage

docs: add comprehensive README with architecture, methodology, and system design

Remove Retrieve Count (Top K) slider component and default top_k to 50 in JS

Update pipeline tracker steps and JS execution to match Multi-Agent Swarm Matrix workflow

Remove pool demographics stats grid from frontend

Empty candidates pool database to clean up all preloaded/inbuilt resumes

fix: resolve tech-biased candidate ranking bug, add healthcare resumes, implement dynamic base SLM loading with cosine similarity & anisotropy centering, and fix keyword pre-filtering phrase-matching bug

feat: integrate specialized sector models and replace COMM model with verified multilingual sentiment model

fix: generalize GeminiGateway prompt and schema to prevent non-TECH sector query bias

feat: conditionally compile model with torch.compile on GPU only

fix: remove torch.compile() — hangs indefinitely on CPU/Windows

fix: replace async/threading with synchronous mega-batch inference

perf: parallelize candidate eval, cache models, optimize inference

Optimize local Qwen VLM loading and thread count on CPU to prevent lockups

fix: Resolve local windows encoding crash, add dynamic Settings panel for HF Token and API Keys

perf: Parallelize Qwen2-VL page OCR extraction, optimize dynamic resolution for fast CPU inference, and flush standard output to prevent perceived freezes

fix: Replace deprecated libgl1-mesa-glx with libgl1 in Dockerfile for newer Debian trixie base images

feat: Implement local layout-aware neural OCR pipeline via Qwen2-VL & EasyOCR fallback

fix: implement native multimodal visual PDF fallback for scanned/textless resumes in ZIP and single uploads

fix: implement dual-extractor (pypdf + pdfplumber fallback) and robust Windows path normalization in PDF/ZIP uploads

fix: Update index.html to read and display specific backend error details for easier Space debugging

feat: Migrate candidate ranking to dynamic GPU-accelerated SLM Swarm evaluation and prepare for Hugging Face Spaces hosting

fix: add HF Spaces README metadata for Docker SDK

feat: add Dockerfile and requirements.txt for HF Spaces deployment

feat: implement SwarmMatrix decentralized multi-agent SLM architecture and relevance optimizations

Initial commit