Use casesSearch & Retrieval
Search and retrieval builds use Jev to match intent, rank candidates, or decide what context belongs in the next step.
This project introduces a novel approach to Retrieval Augmented Generation (RAG) that replaces traditional embedding methods with a structured, table-of-contents-based navigation system, significantly improving speed and reducing costs by offloading non-generative tasks to Jev.
This project rebuilt an AI knowledge graph into a news feed using a decision model for classification, resulting in a significantly cheaper and faster system.
This project integrates TypeSafe AI into a Mac app to provide setup and troubleshooting assistance by reading user questions and referencing the built-in manual to identify relevant documentation or status.
You give it a page and a list of questions, and it sends back a yes, a no or a score for each one, with a probability. 25 judgments about a page take about a third of a second.
Jev is used to filter candidate files for generative LLMs, reducing the number of files sent to the model, tokens processed, and overall cost, while maintaining recall.
Dowse then uses those decisions to select the evidence that reaches the generative model.
Jev is a system designed for fast filtering of news headlines, potentially acting as a preliminary layer before more intensive reasoning by other models.
This project is a launcher that ranks applications by aliases, fuzzy matching, and user habits, with Jev reading user intent to predict and surface the most relevant application.
This project utilizes Jev to enable plain English searches on a job board, allowing users to ask questions directly without needing a predefined schema.
Jev is a tool that reads user intent from keystrokes to provide predictive launching of applications, prioritizing results based on inferred intent rather than traditional ranking methods.
This project scores sentences on a page for meaning using @typesafeai's Jev, highlighting matches.
Atagia, an AI memory engine, is integrating Jev to improve memory-relevance decision-making speed.
This project uses Jev to determine if a web search result for wine is the same as the user's intended query, even with minor discrepancies.
Jev is an AI Assistant reranker that has been deployed to production, showing significant improvements in speed and cost efficiency.
This tool checks the search intent of keywords, utilizing the Jev model for cost-effectiveness and speed.
Messing around with Jev (@typesafeai) and built a grep-like tool that takes plain English instead of a pattern. Every line becomes a noul so you get a probability back per line instead of a match. https://x.com/agentic
Jev is a retrieval mechanism that demonstrated strong performance against pgvector and OpenAI text-embedding-3-small on gold evidence coverage, suggesting its utility for small-to-medium document RAG applications where accuracy is critical.
Jev is a tool that finds relevant repositories, packages, issues, and papers, and then ranks them by relevance and actionability.
A semantic lens built using Jev and LanceDB allows for near real-time semantic filtering without predefining features, enabling search and exploration.
This project implements Jev for search and retrieval, where embeddings find candidates and Jev determines which ones the AI client sees.
