yyouretoast/freightiq

ci: add database seeding step prior to running test verification suite

docs: streamline README overview, eliminate duplicate tool definitions, and refine header

docs: align README guardrails, routing examples, and seeding flags with implementation

fix: resolve SQL relaxer, FTS5 token sanitization, and reranker bottlenecks

docs: add benchmark Pareto tradeoff, category stratification, and trajectory charts

style: standardize retrieval benchmark chart to academic research style

docs(readme): embed native GitHub attachment video player

docs(readme): embed infinite-looping demo GIF with direct video link

docs(readme): embed clean demo video with fallback controls

docs(readme): embed animated walkthrough demo GIF

docs: reconcile Recall@5 benchmark KPI and update asset links

chore(hygiene): remove dead dependencies, align README provider references & default token ceiling

fix(providers): prune dead model IDs, streamline to Groq/OpenAI/Ollama, align token & FMCSA docs

fix(core): patch audit findings (FMCSA gating, SQL relaxation safety, failover isolation, math guards)

fix(tokens): expand output token budget with Groq OTPM safety clamping and conciseness guidance

docs(all): align ADR metrics, docs, env templates, and UI checks with multi-provider architecture

style: remove marketing and ai hype wording across ui docs and benchmark chart

fix(core): remediate CI scoping, benchmark leakage, fallback authority & add multi-provider support

chore(hygiene): remove stale demo.gif attribute and clean up gitignore

fix(compliance): remediate FMCSA hallucination, thread safety, benchmark leakage, and parser robustness

docs(readme): optimize 90-second reviewer journey order, trim routing table, and align one-liner

docs(readme): fix benchmark image LFS CDN resolution and demo video player embedding

style(docs): fix benchmark chart text overlap with dedicated KPI takeaway panel

feat(ui): implement Option C console styling, color-coded routing, and benchmark asset

docs(readme): polish 60-second reviewer journey with honest engineering trade-offs

ci: add automated Hugging Face Spaces sync workflow

docs(audit): synchronize benchmarks, scenario counts, and UI hybrid terminology across all docs

fix(data): replace real company names with 100% fictional carrier identities

docs: restore Representative Query Examples table and format environment variables

docs: remove CHANGELOG.md and prune references

fix: eliminate FMCSA false-positive, harden context windowing, prune dead scripts, and restore LangSmith docs

docs: rewrite README to remove promotional AI hype and align with engineering documentation standards

docs: update README with 500-record scale, 60-query eval, zero-row relaxation & ADR anchors

feat: scale dataset to 500 carriers, expand eval to 60 queries, add SQL relaxation & ADRs

feat(retrieval): implement SQLite FTS5 BM25 + dense RRF hybrid search

fix(app): remove legacy reranker training text, dead session state, and update suggestion chips

Update README and verification suite: remove legacy architecture references, align test counts, and clarify benchmark comparison

Fix NameError in Streamlit tool card rendering by restoring escape and textwrap imports

Refactor README: honest engineering tone, transparent benchmarks, zero buzzword fluff

Elevate documentation to gold standard: Mermaid diagram, trajectory benchmarks, full tool suite, and AGENTS.md

Prune dead code, unused imports, obsolete config, and unify database seeder

Self-contain CarrierReRanker in train_reranker.py for offline experimentation

Modernize FreightIQ: Cross-Encoder reranker, Tavily search, FMCSA verification tool, sidebar API controls, and unified seeder

Update default AGENT_MODEL to qwen/qwen3.8-27b and add automatic 404 model fallback

Fix all audited bugs, architectural design flaws, and performance bottlenecks

docs: update README path references for data/ and models/ directories

fix: loosen requirements.txt version pins for HF Spaces compatibility

refactor: standardize production project layout, data paths, and scripts

docs: clean readme

docs: update README with security validation, rate-limit resilience, early stopping, and known limitations