fix: add retry backoff on transient failures

fix: harden postgres checkpointer connections

fix: update neo4j driver liveness option

fix: staged graph query updates

fix: refactor neo4j driver to singleton to prevent connection exhaustion

fix: use request scoped neo4j drivers

fix: preflight neo4j driver before queries

fix: harden graph retrieval retries

fix: retry transient graph invocations

fix: harden neo4j driver lifecycle

fix: recover neo4j graph connections

update mobile responsiveness

updated docs

frontend: refresh logo and hydration handling

style: refine mobile drawer layout with logo and settings positioning

style: apply interaction polish and mobile spacing

feat: add MobileDrawer and hamburger menu

feat: add BottomNavBar for mobile view switching

frontend: sync hf-safe logo and hydration updates

Merge branch 'main' into hf-sync-temp

fix(api): support Aura Neo4j env names

Merge branch 'main' into hf-sync-temp

fix(api): degrade gracefully when graph is unavailable

Merge branch 'main' into hf-sync-temp

fix(api): avoid asyncio shadowing in lifespan

Merge branch 'main' into hf-sync-temp

topology fix

chore: update cloud DB config and RAG docs (excluding .env)

chore: revert frontend build/serving in HF space as it is on Vercel

fix: build and serve frontend via FastAPI for HF Space support

Merge pull request #2 from adeshboudh/feat/streaming

feat: add logo and fix hydration mismatch

test and docs

feat(eval): RAGAS evaluation framework + RAG pipeline improvements

test(nodes): update patch targets to civicsetu.retrieval.reranker

refactor(nodes): thin wrappers delegating to VectorRetriever and Reranker

feat(retrieval): implement Reranker with score-gap filtering and FlashRank

feat(retrieval): implement VectorRetriever with rrf_merge and retrieve

fix: improve RAG retrieval quality and reduce generator hallucination

fix: deterministic doc_id prevents duplicate chunks on re-ingest

fix: use LiteLLMClient adapter for Gemini judge in build_judge

feat: add Gemini native judge backend to eval pipeline

fix: FTS uses OR-based websearch_to_tsquery instead of AND-based plainto_tsquery

fix: make ingest_all synchronous and fix GraphSeeder method name

fix: correct ingest.py to iterate registry values and call ingest_document

fix: prepend section_title to embedded text so sub-chunks retain section context

fix: graph retrieval fallback uses RRF (vector+FTS) instead of pure vector search

fix: relax reranker thresholds to prevent aggressive chunk filtering

fix: strengthen generator grounding to prevent hallucination on sparse contexts

feat: inject no_reasoning=true into litellm.completion when using osmapi