Blog
& News

Featured

Vector Database Benchmark - OverviewVector databases are crucial for AI-driven applications, but selecting the best one can be tricky. This article breaks down key performance metrics such as recall, nDCG, and QPS, focusing on the HNSW algorithm and VectorDBBench for benchmarking. It also features Docker Compose setups for Milvus, Redis, Chroma, and PgVector to ensure fair and consistent performance testing.
by Luka Panic8 min read
Development
LLM Prompt OptimizationOptimize your LLM prompts for Retrieval-Augmented Generation (RAG) with the JSON-based, and few-shot techniques. The article explores strategies for reducing hallucinations, improving context alignment, and addressing formatting errors. Extensive testing demonstrates the benefits of structuring prompts with clear instructions, separating context chunks, and leveraging in-context learning examples.
by Luka Panic6 min read
Development
Filter
Category
Development (20)Quality Assurance (2)Design (1)
Series
RAG strategies (5)RAG frameworks (4)Vector databases (3)
Topics
Vector Database Benchmark - Chroma vs Milvus vs PgVector vs RedisBenchmark the performance of Chroma, Milvus, PgVector, and Redis using VectorDBBench. This article explores key metrics such as recall, queries per second (QPS), and latency across different HNSW parameter configurations. The results highlight trade-offs in vector search performance. Insights from extensive testing provide a clearer understanding of database behavior, indexing efficiency, and real-world applicability.
by Luka Panic11 min read
Development
A Hot Potato’s Journey to Well Cooked ProductsAs a small but mighty agency, we pride ourselves on our ability to innovate and adapt in the always evolving world of Software Development and Design. At Pixion, our secret weapon for constantly improving our workflow is teamwork and as a Product Designer, one of my favorite methods for working closely with our Engineers to create beautiful and reliable products is the Hot Potato Process. If you haven't heard of it before, let me introduce you to the approach that helps keep our team ahead of the game.
by Katie Tavares6 min read
Design
RAG in practice - EvaluationExplore the RAG evaluation on a large scale to assess the effectiveness of different strategies. This article analyzes answer quality across various configurations, addressing challenges like answer coverage, comparison methods, and rate limits. It also provides insights into the cost and resource impact of running the evaluation pipeline.
by Luka Panic15 min read
Development
RAG in practice - Answer GenerationDiscover how various parameters affect performance in Retrieval-Augmented Generation (RAG). This article explores factors like chunk size, overlap, and context size, examining their impact on the large language model's ability to answer questions. It also delves into error analysis and the cost breakdown of running the full pipeline, including token usage and pricing considerations.
by Luka Panic25 min read
Development
RAG in practice - EmbeddingExplore the embedding process in Retrieval-Augmented Generation (RAG) strategies. This article provides insights into the chunking process, LLM-generated summaries and hypothetical questions, offering valuable analysis on embedding configurations and their impact on answer generation.
by Luka Panic13 min read
Development
RAG in practice - Synthetic Test Set GenerationDive into the practical process of synthetic test set generation with Ragas. This article showcases how AI models are applied to create diverse question types using U.S. Code corpus of data, highlighting real-world testing and cost analysis in RAG applications.
by Luka Panic9 min read
Development
No buzzwords, just greatProduct Development