Build the hero desk objects (skeuomorphic tier 2)

Offer the synthetic example from the hero dropzone

Delete parked Blocks app and claim badges in the Space README

Ship the travel-agency-desk skin and first-run hero dropzone

Add hypermedia presenter for the HTMX custom frontend

Bump gradio to 6.16.0 for gr.Server custom frontend

Default vision adapter to tuned v2 merged model

Show sleepy-mascot empty states on data-less panels

Render the Resumo overview as a row of stat chips

Animate the status panel with mascot state GIFs

Skip extraction for media that already has an artifact

Reuse persisted transcriptions when an export is re-processed

Batch audio transcription and survive transcriber failures

Chunk ZeroGPU audio batches and demote failures to empty text

Add DSPy-aligned v2 vision fine-tune pipeline (T12)

Batch PDF extraction into single ZeroGPU tasks too

Show money as numeric values in the CRM tables

Validate field shapes when coercing extracted values

Teach extraction with few-shot demos and PT-BR classification cues

Give the transformers vision LM a text-only chat path

Extract with the vision LM and JSONAdapter on transformers backends

Pass the extractor to the ZeroGPU worker via fork, not pickle

Batch the whole image extraction into one ZeroGPU task

Harden coercion against small-model noise in reservation fields

Stop the QA fine-tune from looping whole sentences

Show per-kind extraction counts in the dashboard overview

Delete the legacy trip/payment/document model and instructor

Cut the pipeline, dashboard and UI over to MediaArtifacts

Add artifact-based reconciliation, pipeline, identity and summary logic

Add a generic MediaArtifact store to the Repository port

Compose the DSPy artifact extractor and add its deterministic mock

Add per-kind DSPy extraction signatures coercing to domain payloads

Add ArtifactExtractor port and the transcribe-and-classify DSPy stage

Add DSPy and a PortLM adapter over the in-process backends

Add typed artifact payload entities and the ArtifactPayload union

Add ArtifactKind taxonomy and MediaArtifact envelope to the domain

Retry @spaces.GPU calls once on ZeroGPU scheduling errors

Run OCR-only images and PDF pages in one ZeroGPU task

Pass the full chat encoding to generate in the Q&A adapter

Repair missing timestamp tokens in the PT whisper generation config

Flip the Space to ZeroGPU hardware and the CPU llama.cpp wheel

Auto-select backend from detected runtime; add zerogpu-v46 stack

Default the transformers transcriber to the PT whisper-small fine-tune

Add transformers Q&A adapter for the PT-BR MiniCPM5 fine-tune

Extract reasoning-strip helpers out of the llama.cpp adapter

Force llama.cpp to CPU whenever SPACES_ZERO_GPU is set

Add runtime detection module for zerogpu/cuda/cpu

Redact trace payloads at record time; suppress identity documents

Rewrite the Q&A system prompt per Marina's wording

Keep customer facts in the Q&A prompt; trim only old messages