109
Non-autoregressive decision engine on ModernBERT (151M) with calibrated uncertainty (RLCD), TypeSafe AI Jev benchmark audit, and in-browser WebGPU playground
Research and benchmarks — page 2
Benchmarks, papers and independent reproductions. Read these when you need to understand behaviour rather than adopt a library.
537 active projects in this collection. All collections →
106
Personal-assistant agent built on Vercel's eve with 100 mocked tools, measuring how many steps it takes when Jev picks the tool versus the LLM.
97
Agent evals and guardrails as Jev decisions: one request per trace, a fraction of a cent, fast enough for the agent loop. Runs locally with Kev or Laya.
Using Jev as an evaluator.
83
Claude Code skill: hand bulk judgment calls to Jev. 86% fewer Claude tokens on a 12-task benchmark, up to 20x faster. One-line npx install.
72
An open training and inference stack for Jev-style decision models. Train models to score dynamic candidate branches from a shared prefix, with support for high-cardinality choice, calibration, and fast batched inference.
72
JEV-inspired parallel decisions for CUDA LLMs. One context, many decisions. vLLM API, game-agent examples, and reproducible benchmarks.
Persistent local memory for AI coding agents — Claude Code, Codex CLI, Cursor, any MCP client. Temporal knowledge graph, procedural memory, AST codebase ingest, cross-project analogy. LongMemEval R@5 95.1%, LoCoMo 0.607, BEAM 1M 0.448. 74 MCP tools, 9 IDEs, 100% local.
R.A.I.N. Lab is an experimental scientific-agent architecture that separates fast local judgment, independent probabilistic evaluation, multi-agent deliberation, evidence, and authorization into distinct computational layers.🐙(Predates Karpathy's AutoResearch)
62
Community-tested recipes, diagnostics, and benchmarks for CMP 170HX AI workloads
61
An open-source JEV-powered workbench for streaming data selection, quality evaluation, automatic LoRA training and held-out model evaluation.
58
mini-Jev: what a Jev-style typed-decision interface looks like on a frozen Qwen3-4B — read the option letter's logits instead of generating JSON. Preregistered experiment, results, teaching bench.
Open-source alternative to TypeSafe's Jev: a System One style model layer that gives typed, calibrated decisions from any open-weights LLM in one forward pass (HF + vLLM), with honest benchmarks
52
JevGym is an open-source platform designed to benchmark and facilitate better probabilistic estimation in Jev-alike models
Open reproduction of TypeSafe Jev: a 150M typed decision engine (noul/choice/score in one non-autoregressive pass, calibrated confidence). 0.697 vs Jev's 0.727, 2.5x better calibrated, 4x faster, free. Trains on a Colab T4 in 30 min.
45
Turn any off-the-shelf LLM into a Jev -like decision layer
44
an open-source, from-first-principles reconstruction of the ideas behind TypeSafe AI's Jev, written in pytorch
42
Typed JSON inference with DiffusionGemma, with Every and Jev benchmark results
A database-agnostic Agentic GraphRAG framework using swappable System One models (local Laya / cloud Jev). A plug-and-play intelligence layer featuring a complete 4-phase pipeline, continuous evaluation and custom A* traversal for any graph database.
This is a LLM Gateway that mimics typesafe ai structured output. Like an imposter Jev.
37
Replicating Jev with a local LLM
Open replica of TypeSafe's Jev: typed calibrated decisions in one forward pass, on Gemma 4 E2B / Gemma 3 270M (Modal)
35
Self-evolving cognitive decision engine & TypeSafe Jev alternative. Deliberative decision flywheel ('explore first, crystallize later' System 2→1) with 100% option-order invariance. Typed probabilistic API (Choice, Noul, Score) for Open LLMs, Laya (ModernBERT), & commercial Jev with calibrated logprobs and adaptive safety guards.
35
Batched single-token choice inference for open language models, compatible with TypeSafe