Katana

🔥 Sparrow 0.5.0 is out. New model lineup built around three modes: 🚀 Standard — Ministral 3 14B 📊 Tables Only — dots.ocr + Sparrow Templates 🦉 Advanced — Gemma 4 31B Dense Tables Only mode is where Sparrow shines. dots.ocr → HTML → Sparrow Templates → JSON. Reliable structured extraction from complex financial tables, no prompt guessing. Field-level extraction hints. OCR fallback. Agent workflows for non-image data analysis — bonds, financial instruments. vLLM on NVIDIA RTX 6000 Pro. MLX on Apple Silicon. All local. Documents never leave your infrastructure. 5K+ stars on GitHub. Try it: sparrow.katanaml.io Code: github.com/katanaml/sparr…

Large Table Extraction to JSON with dots.ocr — No Vision LLM Hallucinations Sparrow now supports a dedicated table mode for extracting large, complex tables into structured JSON — without Vision LLM hallucinations. Video: youtube.com/watch?v=A7_549… Code: github.com/katanaml/sparr… Sparrow UI: sparrow.katanaml.io/dashboard

New Ministral 3 14B vs Mistral Small 3.2 24B Review I review data accuracy retrieval and inference speed for the new Ministral 3 14B model vs older Mistral Small 3.2 24B. Older and larger 24B model wins this time. Video: youtube.com/watch?v=AxmEp0… @MistralAI #VLLM #Python #Ollama

Ollama and MLX-VLM Accuracy Review (Qwen3-VL and Mistral Small 3.2) Video: youtube.com/watch?v=RotCS3… I was running detail tests to compare accuracy for the same models (Qwen3-VL and Mistral Small 3.2) running on Ollama and MLX-VLM (recent 0.3.7 version). MLX-VLM runs faster, but with lower accuracy. The same is valid across different models. Sparrow: sparrow.katanaml.io Code: github.com/katanaml/sparr…

Comparing Qwen3-VL AI Models for OCR Task Video: youtube.com/watch?v=ZOsg0W… I'm comparing the Qwen3-VL 8B BF16 and Qwen3-VL 30B Q8 models for OCR and structured data extraction tasks. Based on my findings, the quantized 30B model runs faster and with better accuracy than the 8B BF16 model, despite using more memory. Code: github.com/katanaml/sparr… @Alibaba_Qwen

Qwen3-VL Accuracy Differences on Ollama vs MLX Video: youtube.com/watch?v=s6DktZ… I run couple of tests with structured data extraction using newest Qwen3-VL model on Mac Mini M4 Pro with 64GB. I discovered the same Qwen3-VL model with the same level of quantantization performs differently on Ollama vs. MLX. It seems model conversion step is crucial and we must evaluate model performance on different platforms before going to production. @Alibaba_Qwen

Qwen3-VL New Models Comparison and Performance on Mac Mini M4. They are fast 🚀 Sparrow GitHub repo: github.com/katanaml/sparr… Try Sparrow: sparrow.katanaml.io Thanks to @Alibaba_Qwen team and @Prince_Canuma MLX-VLM

Sparrow - my solution for structured data extraction with local AI is growing 🚀 steadily and reached 5K stars 🔥 on GitHub Code: github.com/katanaml/sparr… Sparrow: sparrow.katanaml.io

🚀 Advanced Structured Data Processing in Sparrow This allows to run instruction directly from Sparrow query and check if given fields are present in the document. Video: youtube.com/watch?v=XsGC2r… Sparrow: sparrow.katanaml.io Code: github.com/katanaml/sparr…

Building new pipeline functionality in Sparrow github.com/katanaml/sparr… - AI assistant for code migration from Oracle Forms to APEX. Using Open Source LLMs, such as Mistral Codestral and Qwen Coder. This will run on local machine, completely private, no cloud dependency. @stevemuench

Solving Upwork Client Task with Sparrow I show how Sparrow can be used to handle complex Upwork task, with accurate table data extraction. Key requirement is to prevent Vision LLM hallucinations, this is achieved by Sparrow hybrid data processing approach. Video: youtube.com/watch?v=ZJzAsD… GitHub: github.com/katanaml/sparr… Try Sparrow: sparrow.katanaml.io

Extract financial statement data with Sparrow & Vision LLM! Local processing, no cloud, perfect for private docs. Watch now: youtube.com/watch?v=h_Eblu… Sparrow GitHub repo: github.com/katanaml/sparr… Try Sparrow: sparrow.katanaml.io

🔥 Boost Vision LLM Accuracy with OCR Text Integration I show an interesting approach where I send both an image and OCR text to a Vision LLM 📺 Complete video: youtube.com/watch?v=tlLvPw… 🧑‍💻 Sparrow GitHub: github.com/katanaml/sparr… 🚀 Sparrow UI: sparrow.katanaml.io

I show how Vision LLM misreads "56,000" as "56000" and how combining PaddleOCR’s text extraction with Sparrow’s spatial data processing solves this hallucination issue. Video: youtube.com/watch?v=dHStqS… Sparrow UI: sparrow.katanaml.io Sparrow GitHub repo: github.com/katanaml/sparr…

MCP...

PaddleOCR 3.0: Supercharge Your AI I upgraded to PaddleOCR 3.0 and explain the new PaddleOCR API integration. My goal is to integrate OCR result output with Vision LLM processing to enhance large-scale, structured table data output. Video: youtube.com/watch?v=NqwNIn… Sparrow UI: sparrow.katanaml.io Sparrow GitHub repo: github.com/katanaml/sparr…

New video is out! 🚀 Check out how Sparrow uses the Qwen2.5 vision model to pull bounding box annotations from images, plus add captions and confidence scores. Watch it here: youtube.com/watch?v=kDcLRz… Sparrow: sparrow.katanaml.io #AI #ComputerVision #DataExtraction

Added structured data annotation support to Sparrow, this works with Qwen2.5, not always 100%, but hopefully will improve with Qwen3 VL.

Structured Data Annotation with Qwen2.5 VL and MLX-VLM Qwen2.5 VL can provide bounding box coordinates and confidence values for extracted structured data. This is useful for visual data review and reporting. I will explain with a practical example what prompt should be used to ensure Qwen2.5 returns this data. Video: youtube.com/watch?v=2Ojtc9… Sparrow: sparrow.katanaml.io Code: github.com/katanaml/sparr…