fix(generator): resolve GROQ_FAST_MODEL NameError and empty thought extraction in MarkdownRenderer

fix(rag): prevent Groq TPM 413 rate limit errors with token budgeting and cascade fallback

Clean PDF filename display, format page citations, bump micro font sizes to 12px, add collapsible LLM synthesis in TracePilot, and fix GaugePilot navigation

feat: Context-Aware Glossary, 7 Chunking Strategies, Dual-Tier Memory System, and Structure-Aware Overhaul

fix(prompt): add strict multi-document attribution and sanitize source file headers

feat(retrieval): persist document chunks in PostgreSQL for automatic self-healing across container rebuilds

fix(generation): set max_tokens=4096 to prevent reasoning truncation

fix(retrieval): auto-generate and cache on-demand FAISS index when switching embedding models

fix(benchmark): normalize retrieval_method literals and mount models endpoint

docs: comprehensive README update reflecting 11 query enhancements, active Groq models, and interactive launchpad

fix: align model registry strictly to active Groq models and purge offline routes

fix: add langchain-text-splitters and langchain-core to requirements.txt with resilient native fallbacks

feat: multi-select query enhancements, conversation-scoped docs, rich markdown tables, and UI overhaul

feat: implement multi-document retrieval and pluggable chunking architecture

feat(chunking): introduce modular chunking framework

initial chunking pilotcore

report debug 2

report debug

insights and analysis

Add GaugePilot insights and recommendation engine phase 1

refactors

readme update and gaugepilot custom config fix

...

multi query enhancement

tracepilot pipeline config initialization and gaugepilot expsetup frontend fix

uploading routing for benchmarking

benchmark initial backend and frontend

added additional cross encoder models

fixed bm25

tracepilot trace data and reset traces button behavior decoupled between experimental and production modes, added clear all button to clear conversations/chats from docpilot with decoupled behavior

docpilot chat data decoupled between experimental and production mode, selector behavior fixed

assigning re-ranker its own dropdown

AWS deployment using EC2 and dropdown selector initial mapping

experimentation and production mode frontend

initial benchmark and improve query rewriting

query rewriting initial

ingestion and expire time

feat: improve retrieval evaluation and document ingestion

Added multi model selector capability in chat

feat(metadata): add section-title extraction and trace visibility

improve retrieval evaluation semantics and add reranker lineage observability

consensus frontend

indent

Add sigmoid-normalized reranker confidence & margin, compute retrieval consensus

debug

fix(pilotcore): enable Pydantic v2 validate_assignment on RetrievedChunk to prevent silent lineage field mutation drops

Patch the trace serialization / pipeline assembly layer

feat(tracepilot): add retrieval lineage visualization with per-chunk diagnostics, source badges, and pipeline stage breakdown

feat(retrieval): implement end-to-end retrieval score lineage and observability across dense, BM25, RRF, and reranking stages

add cross-encoder reranking and update README