Remove forced section headers from SYSTEM_WHITE_BOX — /query returns raw LLM output

Parallelize multi-doc retrieval with asyncio.gather

Restore JSON parsing in analyze_document — populate all DocumentAnalysis fields

Append verification warning to answer so frontend always shows it

Speed up analyze: use black_box mode (no thinking, 2K tokens) instead of white_box

Remove _strip_headers post-processing — let LLM output pass through as-is

Strip headers server-side + stronger prompt to enforce plain-text output

Force plain-text output: remove all headers/structure from LLM prompt

Simplify document_agent: drop JSON parsing, return raw LLM output; fix llm.py syntax errors

fix: strict query-only output in analyze — no extra sections

feat: query-driven analyze + context window bump

fix: make analyze JSON structure query-driven instead of rigid template

fix: normalize all LLM JSON fields + wire cross_doc/per_doc

fix: normalize list-typed limitations/methodology from LLM JSON

fix: thinking.type enabled for analyze mode + status event

feat: provider-based LLM profiles + per-request model switching

fix: reduce max_citations_total=10, retrieve_k=15, rerank_top_k=10 + filter chunks <50 chars

chore: white-box temperature 0.75 -> 0.3 (stream + analyze)

feat: remove per-doc citation cap, bump rerank_top_k=20, max_citations_total=50, expose all retrieved citations

fix: retrieve_k=20, strengthen system prompts to prevent citation hallucination, add _sanitize_citations post-processing

fix: black-box concise -> medium reasoning, white-box researcher + analyze -> high

fix: use high reasoning for verify, medium for gaps + full citation text for gaps

feat: add page_width/page_height to citations for frontend bbox scaling + combined bbox+dimension extraction

fix: tune Mercury 2 params — use valid temp range (0.5-0.75), instant reasoning for black-box, high for white-box, reasoning_summary=false for factual paths

fix: use full citation text in _verify instead of truncated 300-char prefix

feat: per-mode reasoning_effort (high for white_box/analyze) + max_tokens 5120

chore: increase white_box max_tokens 3048 -> 4096 for more detailed answers

feat: on-demand pdfplumber bbox extraction for citation highlighting

perf: revert to all-MiniLM-L6-v2, query embed cache, fire-and-forget Cloudinary, pre-warm reranker, reduce top-k

perf: upgrade embedding to all-mpnet-base-v2, add cross-encoder reranker, LRU cache, enable PII by default

production hardening: circuit breakers, graceful degradation, health checks, parallel uploads, HF Spaces deploy config

fix: query and stream_query now greet back instead of analyzing greetings

fix: analyze endpoint no-ops on greetings and when no documents selected

fix: normalize filenames (space/underscore) for doc counting in analyze tab, add document_id to Citation model

fix: analyze tab — documents_analyzed now matches actual citation sources, max_tokens=4096, key_findings 4-7

feat: white-box mode now uses max_tokens=3048 and structured answer output

feat: /analyze now fairly samples all selected documents, adds cross_document_comparison + per_document_summary fields to response

fix: 15 critical/high bugs across codebase

fix models.py: match Neon schema varchar sizes, add email unique constraint