User repository on GitHub
Demo: IA detecta golpe do Pix e rastreia a quadrilha. Rust + Jev vs Python + DeepSeek em tela dividida.
- Forks
- 13
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
Pix Golpe
Split-screen demo pitting Rust plus Jev against Python plus DeepSeek on the same Portuguese scam messages, with fund tracing over a simulated graph of 10 million transactions.
- Stars
- 33
- Language
- Python
- Published by
- Community
- Author
- sandeco
- Last verified
- Oct 2026
About this project
Demo em tela dividida para vídeo. Esquerda: Rust + Jev (TypeSafe AI). Direita: Python + DeepSeek. As mesmas mensagens entram nos dois lados ao mesmo tempo. Cada lado decide se é golpe e, se for, rastreia o caminho do dinheiro num grafo com 10 milhões de transações e 1 milhão de contas. Tudo é simulado. Mensagens geradas por templates, transações geradas com semente fixa. Nenhum dado bancário real. Python 3.12 com websockets, httpx e truststore (pip install -r py-worker/requirements.txt)
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Jev answers four typed questions in one pass while DeepSeek answers the same four as JSON, with latency timed per side.
- Scores at or above 0.6 trigger tracing: 24-hour search from the Pix key, label spread, plus mule and cash-out labels.
- Same tracing logic in Rust and pure Python, checked to match across 40 gangs; graph stays in memory as time-sorted lists.
- Needs Rust, Python 3.12 with websockets and httpx, plus TypeSafe and DeepSeek keys in a local env file.
Watch out
All messages and transactions are template-generated with a fixed seed, and the large transaction file must be generated locally before the race runs. 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.
They come from the author's simulated run with parallel requests per side, not real bank traffic. Regenerate the dataset locally and run both sides yourself to see current latencies and per-side cost.
The tracing algorithm is implemented twice, once in Rust and once in pure Python, and checked to agree across all gangs. The comparison isolates execution time, since both sides answer the same questions on the same data.
Missing API keys are the usual failure. Copy the example env file, set both TypeSafe and DeepSeek keys, generate the data folder once, then start the server. A mock query flag renders fake events with no keys.
No. Messages come from templates and transactions from a seeded generator, with no real banking data. Treat the accuracy table and confusion matrix as a demo harness, not a fraud detector.
You generate it: cargo run --release --bin gerar-dados writes the five files into data/ in about 10 seconds, and because the generator uses a fixed ChaCha8 seed of 42 the output is identical byte for byte, so you can check it against paper/SHA256SUMS. The large transactions binary is not stored on GitHub.
If the scam probability is at or above 0.6, it runs a 24-hour temporal search from the Pix key, scores accounts for suspicion, propagates labels, and classifies mule and cash-out accounts. The same algorithm runs in Rust (src/graph.rs) and pure Python (py-worker/trace.py), verified to match bit for bit across 40 gangs.
Jev answers four Portuguese typed questions in one pass: is it a scam, what type, how urgent, and does it ask for a Pix. DeepSeek answers the same four as JSON, and latency is measured on each side. Every message and transaction is template- or seed-generated, with no real banking data.
More Jev resources like this
46GitHub 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/TypeLLMPython45GitHub stars
Drop-in service that exposes Upstage Solar models through the System One shape, accepting state plus noul, choice, or score questions.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
hunkim/solar-mini4-jevPython
hunkim/solar-mini4-jevPython38GitHub stars
Research recipe that fuses two frozen open readers into Jev-style yes/no judgments with a fit-free router, confidence gate, and candidate sets.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
FeiLiuEM/open-medical-jevPython
FeiLiuEM/open-medical-jevPython62GitHub 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
40GitHub stars
Open research toolkit that encodes an image once and answers multiple typed choice, noul, and score questions from a shared distribution, plus a local 0.6B text-decision model.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
IamBusy/OpenJev-VisionPython
IamBusy/OpenJev-VisionPythonFrom 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.