MachineLearning-Nerd/jev-tetris
User repository on GitHub
A visual TypeSafe demo where Jev chooses verified Tetris placements.
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
Jev Tetris — MachineLearning-Nerd
A Streamlit Tetris where the engine enumerates every legal placement, Jev picks one typed placement ID, and Python verifies rotation, column and exact occupied cells before the piece is allowed to lock.
- Stars
- 1
- Language
- Python
- Published by
- Community
- Author
- MachineLearning-Nerd
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is Jev Tetris — trungdq88.
About this project
This is a public, company-data-free TypeSafe experiment. It demonstrates a useful pattern for Jev: give the model a finite set of valid outcomes, ask it to select one typed option, and let ordinary code verify and execute the choice. Jev is not asked to invent keyboard commands or output arbitrary coordinates. The engine enumerates the legal final placements for the current piece. Each candidate contains: rotation (0, R, 2, or L); leftmost occupied column and landing row;
Quoted from the project's README. Read the full README.
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Each candidate carries rotation, leftmost occupied column, landing row, exact occupied cells, holes, height and bumpiness.
- The answer is rechecked against the board before every lock: if the placement ID is no longer legal, or its cells do not match, the move is not applied.
- Two play modes: Ask Jev for Placement advances one piece, Jev Playing Tetris repeats the loop until paused or game over.
- A deterministic local planner runs the whole loop without credentials, and its traces are labelled previews and never sent to Jev.
Watch out
The model's job is narrower than playing Tetris: the engine enumerates the legal final placements and Jev chooses an ID from that finite set, so the ceiling on play quality is set by the enumeration, not by the model. Two things will mislead you if you are not careful. The deterministic local planner stands in for Jev whenever no key is set, so an unconfigured run looks like the model playing and is not. And live mode requires TYPESAFE_API_KEY in the server environment, where it is read by Python and never sent to the browser, with .env.example documenting the variable without holding a secret. The provided checks are unittest plus py_compile over the engine, the placement contract and the app modules, which cover physics and enumeration rather than live API behaviour. The board is zero-indexed with columns 0 through 9 and rows 0 through 19 from the top, and the candidate column is the leftmost occupied cell, which matters if you read a JSON trace. Earlier synthetic Agent Courtroom and SQLite experiment files remain in the folder for comparison even though the Tetris playground is the default app. All state is synthetic and company-data-free. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
With a key set, Jev returns a placement ID and the engine verifies rotation, column and exact cells before locking the piece, so the model cannot hallucinate a move. Without a key, though, a deterministic local planner substitutes and the README labels those traces as previews that are not sent to Jev, which is the honest distinction: the same interaction is available offline and none of that offline play says anything about the model.
It never sees pixels. The engine enumerates the legal final placements for the current piece and packages each with rotation, leftmost occupied column, landing row, exact occupied cells, lines cleared, holes, height, bumpiness and a one-piece lookahead score, then Jev selects one placement ID such as p11. That is the pattern the project is demonstrating: a finite set of valid outcomes, one typed choice, verification by ordinary code.
Yes, and it is the most useful part for learning the API. Two collapsible panels show the JSON input, state and question, and the JSON output, Jev's response, and Export JSON trace downloads the complete decision history for the current game including each selected placement, confidence, probabilities and verified result. Live mode also shows confidence and the alternative probabilities as they arrive.
In the server environment, exported before Streamlit starts: export TYPESAFE_API_KEY, then uv run streamlit run app.py. .env.example documents the variable without holding a secret, and the key is read by Python and never sent to the browser. Setup itself is uv sync, uv run streamlit run app.py and localhost:8501; the provided checks are unittest discover plus py_compile over the engine, the placement contract and the app modules.
The engine enumerates the legal final placements for the current piece, Jev selects a placement ID, and the engine checks the ID is still legal and its exact occupied cells match before locking the piece.
SRS-style rotations with wall and floor kicks, 7-bag piece generation, a five-piece next queue, and a ghost landing preview, with animated rotation, movement, drop, and lock.
Export JSON trace downloads the whole game's decision history with each selected placement, confidence, probabilities, and verified result; local-planner traces are labeled previews and are not sent to Jev.
More Jev resources like this
4GitHub stars
Browser-use layer for CLI coding agents: Jev picks the operation and target element in one round trip per step, exposed over MCP with a CLI, a Python API and an HTTP server.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
brnyxx/jev-raPython
brnyxx/jev-raPython0GitHub stars
A local Python browser automation UI that turns a URL and a plain-language goal into bounded actions, with Jev picking the next operation from observed candidates and a separate text model used only to fill fields.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
KesavanKing/jev-browserPython
KesavanKing/jev-browserPython46GitHub stars
Open 150M encoder that reproduces the typed noul, choice, and score interface in one forward pass, trainable on a free GPU and runnable locally for nothing.
Jev Demos & Experiments#community#jevbest#catalog-import
941GitHub stars
Library that adds schema-guaranteed typed outputs to existing autoregressive LLMs without changing weights, served through SGLang or a hosted API.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
TypeLLM/TypeLLMPython
TypeLLM/TypeLLMPython62GitHub stars
Simulation study that drives one xArm7 apple-to-plate task with Jev, a large reasoning model, and a small model through identical incremental controls.
Jev Demos & Experiments#community#jevbest#catalog-import
39GitHub stars
Mac menu-bar voice controller where Jev picks the next action from observed Accessibility controls and Swift executes it in an observe-decide-execute loop.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
ronadin2002/jev-cuaSwift
ronadin2002/jev-cuaSwiftFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.