update live space url

Refactor README and BLOG documentation for clarity and improved structure

Update README and app.py to reflect model organization changes and improve clarity in documentation

Update evaluation documentation to reflect post-fix results and clarify judge methodology

Refactor code structure for improved readability and maintainability

Refactor code structure and remove redundant functions for improved readability and maintainability

Add table borders to PDF export for improved visual separation

Update market facts statistics and sources for accuracy and relevance

Enhance scenario and portfolio JSON generation with randomized context and messaging for improved variability in outputs

Enhance user experience by adding error handling for revision actions, improving content reflection in drafts, and introducing a baseline LLM client for fallback generation

Refactor revision action UI: move whole-document revision to Work Sample tab and update button styles

Enhance landing screen with structured five-step journey and improved styling; update demo profile prompts for clarity

Enhance user flow by improving dropdown behavior and adding clear call-to-action elements

Refactor artifact selection logic and enhance user experience

Split Export onto its own tab; five-step progress rail

Give the full work sample its own Work Sample preview tab

Drop duplicate sidebar progress card (top stepper covers it)

Build step: split editor into columns (nav | answer | draft)

Compact layouts: two columns across all steps, trim intro/hero/fields

Build step: true side-nav — show only the selected section, collapse full doc

Add README screenshots, submission blurb, and example case study

Add UI screenshots of the new 4-step flow + capture script

Simplify UX: 4-step flow, live Build view, split-artifact export

Add insights on model fine-tuning and performance differences between direct SFT and distill-then-quantize approaches in documentation

Refactor fiction penalty logic in review service to improve accuracy; replace broad match with specific negative and reality checks for critiques.

Enhance PDF export functionality by flattening table cells and improving error handling; update review prompt for detailed feedback and adjust strength/weak spot limits.

Update model description for distilled Qwen 0.5B to clarify source

Enhance functionality by adding CJK character filtering, improving prompt detail handling, and updating dependencies

Refactor portfolio and scenario services to utilize prompt formats for consistency and clarity

Merge pull request #4 from visprogithub/hf-hackathon-model-perf

GGUF threads: prefer physical cores locally (avoid SMT oversubscription)

Bind launch to 127.0.0.1 locally (0.0.0.0 only on Spaces)

README: correct functionality / architecture / run-locally sections

Remove template/heuristic fallbacks: surface real model output only

Constrain and validate JSON sub-tasks so small models stay reliable

Hard-code fine-tuned model dropdown; speed up both fine-tuned models

fix: handle invalid JSON structure in complete_json method

feat: add JSON-compliant completion method to LocalLlamaCppBackend

feat: enhance model choice display in app.py for clarity

fix: clean up README by removing unnecessary whitespace and improving sentence structure

Merge branch 'distill-adapters-smaller-model'

Add evaluation analysis and judging scripts for ProofKit models

Add comprehensive specifications and user stories for ProofKit

chore(kb): periodic knowledge base refresh

feat: Add documentation for agent-assisted build process and enhance .gitignore for new files

feat: eval + charts compare the 6 requested models incl. distilled via llama.cpp

feat: scale local Qwen2.5-0.5B notebook for the 7000-sample retrain

feat: distill fine-tuned gpt-oss-20b to a small model for free-Space hosting

feat: Enhance evaluation process with detailed chart generation and checkpointing for training runs

Refactor fine-tuning scripts and documentation for ProofKit