Fix Docker build OOM; fix missing bm25_prefilter in Streamlit; update docs

Add bm25 prefilter stage; unify candidates cache; validate reasoning quality

Run JD parse sequentially; reduce parse_max_tokens

Reduce top_n defaults for embedding & cross-encoder

Refactor JD extraction: relocate artifact & optimize parsing

Switch JD-parse/T1 polish to Qwen3-1.7B

Respect --limit; LLM streaming heartbeat

Add JD to image; logging banners; safe explain

Add LLM call timeout and streaming pipeline phases

Add observability logging for LLM and embedder

Add timing and error logging throughout

Cache-first model loading for zero-network ranking

Add CPU threading, T1-only LLM polish, experiment script

Remove char limits; expand template variants

Reuse Qwen3.5 for T2; drop LFM2.5 & factor ONNX

Explain: randomize the over-used 'Built' opener across build-synonyms

Explain: research-based outcome notes (prompt-engineered + verified)

Explain: natural 'relevant to the requirement' tail, high variety

Explain: A+B sentence structure, behavioral framing, more variety

Explain: revert LLM polish that collapses to <3 sentences

Streamlit: write submission.csv as the final pipeline stage

Models load fully offline + enable runtime auto-download

Docker: pre-download all models at build for offline ranking + docs

Multi-band explainability: tiered NLG + dual-model polish (Step 11)

Enhance JD extraction (Qwen3.5-2B + richer prompt) and Step-11 reasoning

Enrich the granite-embedded candidate text with familiar skills + pillar strengths

Enrich the embed/rerank query with extracted skills + roles

Explainability: true-random phrasing + drop dead reasoning module

Add JD-grounded explainability for the top 100 (Step 11)

Use the JD locations field: add a field-scoped BM25 locations group

Tune JD parse: generic compressed prompt, best-of-N recall, JD phases in explorer

Add field-scoped BM25 and composite blending

Streamlit explorer: inputs, metrics & report

Drop disqualifiers field — remove dead JD signal

Update Dockerfile and refine JD parsing prompts

Fix JD OBJECTIVE prompt: request all fields the funnel scores on

Prefer prebuilt llama-cpp-python wheels

Add reasoning LLM, perf tweaks, and UI timing

Add full pipeline, submission & reasoning support

Add ensemble ranking step

Add ONNX granite/reranker and compute config

Add retrieval & soft/composite ranking stages

Add JD parsing, central config, checkpoints & LLM

Ship data in image + orjson loader; fix HF file-not-found

Updated

update

Duplicate from streamlit/streamlit-template-space