Jev Development Demos

Reading the current Claude Code task and recent context to select the best-fitting model automatically

Building a blog, creating a post, and opening shared items from spoken instructions alone

Highlighting code through an open-source library with no external dependencies and support for Vercel and TypeSafe gateways

Organizing a knowledge base monthly, classifying each item automatically, and alerting a person only when judgment is needed

Adding typed decision endpoints and multimodal input to existing decoder-only models such as Qwen and Gemma

Optimizing an existing workflow until its end-to-end runtime reaches 56 seconds

Reading every reply in a Chrome extension and hiding those classified as low effort

Selecting a sub-agent model, role, and reasoning effort from a task brief and eligible worker profiles

Searching 1,783 shadcn components by six semantic dimensions and returning the matching install command

Removing the full Claude Code skill list from context and loading only the single skill most relevant to the request

Turning every agent choice into a confidence-scored answer and using a code gate to act, hold, or ask a person

Searching Kubernetes logs semantically from the terminal or dashboard with queries such as database issues or source IP

Combining Whisper transcription with typed intent decisions to control Omarchy Linux by wake phrase or F10

Turning plain-English code-issue descriptions directly into live dashboards grouped by owners, folders, and paths

Testing a small typed-decision prototype and then integrating the pattern into several existing projects

Routing simple coding tasks to cheaper models and difficult work to stronger ones while displaying credits used

Judging 150 code comments in 9.3 seconds for about one cent and converting probabilities into plain conditional rules

Highlighting CUDA kernel code with likely stride and performance issues before compilation or profiling

Checking a live architecture canvas against explicit design policies and surfacing rule violations as the model changes

Choosing model and reasoning effort automatically for Codex and Claude Code instead of defaulting to the strongest option

Reviewing generated UI against approved components and Figma designs, detecting AST drift, and feeding fixes back into the coding loop

Selecting shadcn components and style rules to assemble a responsive interface in one to two seconds

Reranking existing site-search results instantly through a drop-in React component

Validating form fields against requirements written in natural language