SYSTEM CONNECTING - 0 cache
Knowledge that makes AI smarter
NEXUS is private RAG infrastructure. Parent child chunking, hybrid RRF, MMR diversity, semantic cache 0.93. Works with any AI model, even 128M. Light default, dark auto, pure HTML, no lag.
768d IndoBERT-~120ms-no Streamlit
Parent 1500 chars Child x3 500 + overlap Vector 768 dim Context Ready for LLM
Nexus Engine 1.5 Pure
Faster, smarter retrieval, zero lag dashboard
No Streamlit, no extra_streamlit_components, no iframe bug. Pure HTML JS with Tailwind-like CSS, batch embedding, hybrid RRF, MMR.
Hybrid Search
Dense 40 + keyword + RRF k=60
50ms
Rerank Batched
Cross encoder batch 32, 0.7+0.3 fusion
30ms
MMR Diversity
Lambda 0.7, diverse parents
5ms
Semantic Cache
Threshold 0.93 LRU 500
8ms
Synthesis
Compress, dedup, join ---
15ms
Indonesian First
LazarusNLP IndoBERT 768d for Bahasa Indonesia slang and formal.
Parent Child Semantic
Recursive by sentences, not fixed chars. Child 500 for search, parent 1500 for context.
Hybrid RRF MMR
Reciprocal Rank Fusion k=60 plus MMR lambda 0.7 for diversity.
Zero Lag Pure HTML
No Streamlit, no CookieManager bug. FastAPI serve static, all API proxied.
Zero Cost
Runs on HF free tier, Qdrant free 1GB, no fine tuning needed.
Private Secure
Bearer dashboard token, RAG API key hidden server side, sliding window 3000/h.
0
Parent Chunks
0
Child Vectors
0
Queries Today
NEXUS - Neural Extensible eXtraction with Unified Synthesis
v1.5.0 pure html - light default dark auto