Developer Ditches Embedding APIs for Free Local CPU Solution Using FastEmbed

·1 views

A developer building a Retrieval-Augmented Generation (RAG) side project abandoned it after finding the reliance on paid third-party embedding APIs — first Gemini, then Mistral — impractical for personal learning. Returning to the project months later, he discovered FastEmbed, a lightweight library that generates text embeddings locally on a standard CPU without requiring a GPU or heavy dependencies. He adopted the BAAI/bge-small-en-v1.5 model running on ONNX Runtime, which handled the workload on his ordinary laptop without performance issues. The revised architecture also dropped the assumed need for a dedicated vector database, instead using plain Python data structures and NumPy cosine similarity calculations for document retrieval. The developer noted that simplifying the stack — removing external APIs and specialist infrastructure — gave him a clearer understanding of how RAG actually works.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Browser-Based PCB Tools Are the Only Viable Option for Mobile and Chromebook Users

Traditional desktop EDA software like KiCad, Altium, and Eagle cannot run natively on Android, iPadOS, or Chromebooks, leaving mobile users with viewer-only apps that do not support editing. The one common capability across all these devices is a modern web browser, making browser-based tools such as EasyEDA and Flux the practical path to mobile PCB design. Chromebook users face the sharpest constraint, as they cannot install desktop EDA at all, making a browser-based editor their only functional option. Touchscreen limitations make manual component placement difficult on small screens, but AI-assisted tools can help by letting users describe a board in plain language and auto-generating placement and routing. Experts caution that reliable tools should separate AI-driven design decisions from deterministic geometry engines to ensure outputs meet fabrication standards.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

IBM Study: AI Agents Passing 77% of Tests May Succeed Only 53% in Real Use

IBM Research has identified a 'consistency gap' in AI agent performance, revealing that a ReAct agent using GPT-4.1 achieved a 77% per-run pass rate on the AppWorld benchmark but completed the same task successfully across all five repeated runs only 53% of the time. The 24-point gap arises because small variations in early steps of an agent's reasoning loop compound across subsequent steps, producing different outcomes for identical inputs — even without any code, prompt, or model changes. To address this, the researchers built a Consistency Analyzer that identifies the exact step where agent trajectories diverge, then generates targeted guidelines stored as episodic memory for future similar tasks. This approach raised the all-five pass rate by 16 points on known tasks and by 13 points on unseen but similar tasks. The findings challenge standard AI testing practices, where each test case is typically run only once, potentially masking significant real-world unreliability.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Why Security Teams Must Add Context to Internet Exposure Data

Internet exposure counts are commonly used by security teams to assess attack surfaces, but raw numbers can be misleading without proper context. A single machine may expose multiple services, meaning service counts and device counts are not interchangeable and should not be treated as equivalent. Identifying a product on the internet does not confirm it is vulnerable, as affected versions and specific configurations must also be considered. Exposure data should always include the collection timestamp and clear criteria used to identify the product, enabling more accurate comparisons over time. Analysts are advised to clarify the counting unit, the data source, and whether results indicate exposure or confirmed vulnerability before presenting findings.

0 comments Read more at DEV Community

ProgrammingHacker News ·

Review: How Greek Shaped the Bible and Early Christian Thought

A review in the Los Angeles Review of Books examines Timothy Michael Law's 2013 book 'When God Spoke Greek,' which explores the role of the Greek language in shaping religious scripture. The book argues that the Greek Septuagint — an ancient translation of Hebrew texts — had a profound influence on early Christianity. The review discusses how this Greek version of scripture was central to early Christian writers and theologians, often overshadowing the original Hebrew texts. The piece highlights the broader cultural and theological implications of language choice in the formation of religious tradition.

0 comments Read more at Hacker News