by Kashish

Building a Validation Layer for Financial Data in LLM Pipelines

Even after applying model optimization and output cleaning, one critical issue remained in the ReceiptFlow pipeline: […]
read more
by Kashish

Debugging LLM Output in Real-World Receipt Processing

Even after selecting the right model and optimizing input format, inconsistencies in LLM outputs continued to […]
read more
by Kashish

Input Representation vs Model Performance in OCR + LLM Pipelines

After selecting an appropriate model for structured extraction, inconsistencies in output still persisted within the ReceiptFlow […]
read more
by Kashish

Best Qwen Model for Receipt Extraction (0.8B vs 3B)

After identifying that tool calling was unreliable in a local LLM setup, the next critical step […]
read more
by Kashish

Why Tool Calling Failed in llama.cpp (Qwen 2.5)

When building ReceiptFlow, the goal was simple: take messy OCR output from receipts and convert it […]
read more
by Chris

Sovereign Enterprise AI for Data Analysis with Apache Druid and Clickhouse

This article explores the concept of Sovereign AI and introduces Data Philter, an open-source, local-first AI […]
read more
The title image is a visually engaging graphic (1200x628 pixels, per Search Engine Journal) depicting a stylized, simplified version of the data ingestion pipeline flowchart from the article. It features a clean, dark blue background (#005580) with a central flowchart of 6 steps, arranged vertically, using transparent rectangles with black borders for steps (e.g., “Source Crawling,” “Data Preprocessors”) and white document shapes for outputs (e.g., “Chunked Data”). Black ellipses highlight processing stages (e.g., “Data Embedding”), connected by black arrows with arrowheads. Dashed containers list extensions (e.g., “Vector Embedding Generation,” “Filter Specification”) in Arial font (12px for steps, 11px for extensions). A subtle overlay of interconnected nodes and data flow lines (representing polyglot databases like vector, SQL, graph) spans the background, with icons for text (document), images (photo), and JSON (code brackets). The article title, “Scalable Polyglot Data Ingestion Framework for AI-Driven Search Ecosystems,” is overlaid in bold, white Arial font (24px) at the top, with a tagline “Enabling Vector, SQL, and Graph Indexing” in smaller text (16px) below.
by Tim

Polyglot Knowledge RAG Ingestion Concept for Enterprise Ready AIs

Explore a scalable polyglot data ingestion framework for AI-driven search ecosystems, supporting vector, SQL, and graph indexing. A flowchart details 6 steps for preprocessing and embedding, enabling robust RAG search.
read more