DEDUPEDScientific Agent Skills
A library of 177 Agent Skills that give AI agents documented, validated guidance for scientific packages, databases, and research workflows.
4 use cases: Pull compound and protein records with source provenance; Screen a compound list for drug-like properties; Run a single-cell RNA-seq analysis with Scanpy · GitHub
Speaker Mem R1
SpeakerMem-R1 is a Python research repo with a speaker-centered memory package, benchmark runners, writer-training recipes, and evaluation outputs for multi-party dialogue.
4 use cases: Retrieve speaker-labeled evidence from a group chat locally; Benchmark BM25 baseline retrieval on GroupMemBench; Evaluate SpeakerMem on SocialMemBench with a local Writer · GitHub
DSH Council
A DSH Council plugin that asks several AI models to answer a question independently, review each other's answers anonymously, and produce one inspectable final recommendation inside DeepSeek Harness.
3 use cases: Stress-test an architecture decision before committing; Map where AI models agree and disagree on a research question; Check a business plan for blind spots before deciding · GitHub
Evo Bot
EvoBot is a Python agent and reusable skill library that runs manipulation tasks in the LIBERO robot benchmark, reporting 25 of 30 tasks passed.
4 use cases: Reproduce the 25 of 30 LIBERO evaluation result; Watch one LIBERO task learn across retry attempts; Resume the LIBERO agent runner and skip passed tasks · GitHub
AI Engineering Course
A free, step-by-step AI Engineering course in Markdown, with 18 modules and 146+ lessons from machine learning basics through LLMs, RAG, agents, inference, evaluation, safety, and system design.
4 use cases: Map a step-by-step learning path into AI engineering; Prepare for AI engineer interviews with course material; Compare LLM serving tools before picking one · GitHub
AI System Design
A Markdown guide that teaches AI system design step by step, covering LLM inference, RAG, AI agents, MCP, guardrails, evaluation, and interview preparation.
4 use cases: Prepare for an AI system design interview; Plan a RAG architecture for document Q&A; Design an AI agent with tools and memory · GitHub
LLM Inference Engineering
A free, step-by-step learning series of blog posts and videos on LLM inference engineering, covering topics such as KV cache, PagedAttention, continuous batching, and speculative decoding.
4 use cases: Learn how LLMs generate text one token at a time; Pick a KV cache memory technique for long prompts; Choose a faster decoding method for slow responses · GitHub
Open Ghost
OpenGhost is an open desktop AI agent for Windows, macOS and Linux that runs commands, edits files, works in a built-in browser and draws explanations beside its replies.
4 use cases: Explain a topic with drawings beside the text; Review your account activity through the agent's browser; Save notes the agent raises when the topic comes up · GitHub
4DAnyone
4DAnyone is a Python research repo that turns a casual monocular video of one person into consistent multiview videos for downstream 4D Gaussian Splatting reconstruction.
3 use cases: Turn a casual clip into a 24-view orbit set; Preview camera layouts in a browser GUI; Run a quick 6-view test on a consumer GPU · GitHub
YouTube Skills
A set of 12 Agent Skills that let coding and chat agents pull YouTube transcripts, search YouTube, read channels and playlists, and run bulk exports through a hosted API.
4 use cases: Summarize a YouTube video from its transcript; Bulk export a channel's transcripts for a search index; Find every mention of a topic inside one channel · GitHub
Claudemy
Claudemy is a free, open-source Claude Code skill that researches a topic on the web and builds a course of about 12 two-minute audio lessons with cited sources.
3 use cases: Build a 12-lesson audio course on any topic; Build a course from one expert's blog or Substack; Map where experts disagree on a topic, as cards · GitHub
LLM Legislativo
A Brazilian legislative text corpus from the Câmara dos Deputados and Senado Federal, plus a causal Transformer language model trained from scratch in pure PyTorch.
4 use cases: Study a from-scratch causal Transformer in PyTorch; Load the Brazilian legislative model and generate text; Build a Portuguese legislative text dataset · GitHub
Robo Harness
RoboHarness is a Python harness that lets LLM agents control simulated robot manipulation tasks in BEHAVIOR through a visual-geometric control panel, using tracked image keypoints and depth back-projection.
3 use cases: Reproduce a BEHAVIOR task mean Q-score; Watch one agent rollout in a browser; Point the harness at your own model endpoint · GitHub
Bootloops
BootLoops is a Python harness of certified scientific computing tools and house engines that LLM agents can run for exact and high-precision physics and quantitative science.
4 use cases: Validate a fresh BootLoops clone before trusting it; Reproduce the one-loop box with the Landau Alphabet engine; Have an agent map a precision problem to the right tools · GitHub
Dynolab
Dyno Lab is a free, open-source Mac app for running local AI models and investigating their behavior with activations, probes, SAEs, interventions, agent tests, and a Python SDK, HTTP API and MCP bridge.
5 use cases: Run Claude Code or Codex on a model you own; Test whether an agent breaks a rule to reach a goal; Reproduce the identifier-fidelity probe experiment · GitHub
JEV Sees
JEV Sees is an open-source Python project that detects objects in images, video, and RGB-D camera feeds, tracks them across frames, and asks the JEV model a question about each one.
5 use cases: Ask JEV a color question about each object in a photo; Score each tracked pedestrian's accident probability in a clip; Label each tracked animal's motion state in a video · GitHub
Textfold
TextFold is an open-source Python tool that folds papers, web pages, files, and AI replies into five levels of detail, from a one-screen overview down to the original text.
5 use cases: Read a long paper from overview to full text; Let an agent read a document map first, then one section; Fold long Claude Code replies to a level you can scan · GitHub
Orca Bonsai 27B Uncensored
A runtime refusal-direction ablation layer that changes the behavior of the bit-identical Ternary Bonsai 2 27B model without modifying or re-quantizing its weights.
3 use cases: Compare refusal behavior with and without ablation; Confirm the ablation hooks fire before evaluating; Sweep alpha values to map partial ablation strength · GitHub
LLM Wiki V2
A Python library and web app that turns your documents into a Markdown wiki of notes and answers questions with citations to those notes, using any LLM provider.
5 use cases: Turn a pile of documents into a cited Markdown wiki; Build a wiki that runs on your own computer with Ollama; Run the wiki as a server that other tools can call · GitHub
Mini Verl
miniVERL is an independent Apache-2.0 Python project that runs verl-style PPO, GRPO and on-policy distillation training on one NVIDIA consumer GPU.
4 use cases: Run a first PPO experiment on one GPU; Run an existing verl config on one GPU; Distill a teacher model into a smaller student · GitHub
Indra
Indra OCR is a CPU-based OCR tool that turns PDFs and images into markdown, text, or JSON with boxes, layout blocks, and reading order.
4 use cases: Turn a folder of scanned PDFs into markdown files; Get OCR lines and boxes as JSON for a data pipeline; Check OCR results on each page in a local viewer · GitHub
Open EBFM
OpenEBFM is open-source Python code for training, scaling, evaluating, and chatting with energy-based foundation language models.
4 use cases: Pretrain an EBFM model to a chosen token budget; Chat with the published EBFM SFT model in a browser; Fine-tune a published EBFM checkpoint with SFT · GitHub
Zero to Sglang
A free, open course from Datawhale and RadixArk that teaches LLM inference by building a mini-sglang from scratch and then reading the real SGLang source.
5 use cases: Learn LLM inference concepts without a GPU; Build a mini inference engine step by step; Run Qwen3-0.6B on SGLang on your own GPU · GitHub
Donsetch
DonSeTch is a Rust binary that gives AI agents web fetch, keyless search, crawl and screenshot tools through MCP or a standalone CLI.
4 use cases: Give a coding agent live web search and fetch; Read only the needed part of long web pages; Pull text out of scanned PDF reports · GitHub
STEP Quant
STEPQuant is Python code that compresses the persistent recurrent states of Delta-rule models by deciding which state units get more bits and how keys and values are scaled, with SGLang serving support.
3 use cases: Reproduce STEP Quant@6 results on Qwen3.8-27B; Serve a Delta-rule model with compressed recurrent state; Compare FP32 and quantized state accuracy on long tasks · GitHub
EMERGE Policy
EMERGE-Policy is a Python embodied-agent framework where a primary agent plans and calls specialized perception, action, and verification models to control robots in LIBERO simulation.
4 use cases: Run a robot task from a browser console; Run one robot task headlessly and get JSON back; Reset and switch simulation scenes from the terminal · GitHub
Awesome AI for Games
A curated, searchable collection of research papers on AI and foundation models across the game lifecycle, organized into six roles.
4 use cases: Build a reading list for LLM game NPC agents; Find benchmarks for evaluating LLM agents on games; Map AI-for-games research across six lifecycle roles · GitHub
Kimi K3 in C
A C99 inference engine that runs the 2.78-trillion-parameter Kimi K3 model on a single CPU with no GPU, streaming weights from disk when RAM is small.
4 use cases: Run Kimi K3 text completions on an 8 GB laptop; Chat with Kimi K3 offline and save the transcript; Check that output stays identical across memory presets · GitHub
AI Learning Guide
A desktop and browser app with 51 hands-on labs that teach how modern AI works, from ML basics to transformers, RAG, agents, fine-tuning and quantization.
4 use cases: Learn AI basics through 51 labs in a browser; Run the desktop app locally for a study group; Add a new lab to the Discover AI app · GitHub
Carrel
Carrel is a self-hosted knowledge base that parses, indexes, and searches your documents so AI agents can retrieve source passages, images, and related facts through an API.
4 use cases: Let a coding agent search your product docs; Answer contract questions with cited passages; Trace decisions and conflicts across meeting notes · GitHub
Flybrain Robot Bridge
A Python experiment that connects a camera, a small hand-designed fly-inspired neural model, and a UDP link to a physical robot, with a dry-run mode by default.
3 use cases: Run the synthetic fly-brain demo without hardware; Test motion and looming output from a webcam or video; Calibrate a physical robot link in safe steps · GitHub
Jevbench
JevBench is a benchmark and leaderboard for AI decision models that measures typed decision accuracy, calibration, latency and cost.
3 use cases: Shortlist decision models from the JevBench board; Test a decision model for option-order sensitivity; Compare local decision models with stuntdouble · GitHub
Anima
Anima is a Claude Code skill that turns a biology idea into a standalone, interactive, sourced isometric HTML figure.
3 use cases: Make a sourced, interactive biology figure from one prompt; Adapt a bundled example into your own biology figure; Check a figure zoomed in, in both themes and at phone width · GitHub
State of Thought
State-of-Thought is a Python release of a 582-parameter controller that reads frozen language and vision-language model states to decide when to retrieve evidence and when to stop reasoning.
5 use cases: Audit a released SoT controller checkpoint; Check published results against bundled records; Evaluate the frozen controller on a text benchmark · GitHub
Scientific Illustrator
Scientific Illustrator is a Codex plugin that redraws a reference figure as editable objects in PowerPoint, WPS Presentation, or draw.io, then checks and corrects the result.
3 use cases: Redraw a figure in draw.io and export a 2000 px PNG; Redraw a figure in WPS Presentation as an editable PPTX; Watch figure objects draw live in Mac PowerPoint · GitHub
Academic Research Skills
A set of Claude Code skills that cover academic research from planning and literature review through writing, peer review, revision and finalizing a paper.
4 use cases: Map a paper's chapter structure through guided dialogue; Produce a literature review on a topic in one command; Import the five ARS skills into Claude Science · GitHub
Test Model Thing
A small proof-of-concept byte-level language model (not an LLM) built with MLX that explores latent-space prediction and recurrent trace units.
3 use cases: Train a small byte-level model on your own machine; Chat with a frozen model to inspect byte output; Benchmark your model on CoLA against GPT-1 · GitHub
Auto WWTP ASM Demo
AutoWWTP-ASM is a Python LLM agent workflow that builds and calibrates ASM models for wastewater treatment plants from a task text and spreadsheet data.
3 use cases: Calibrate an ASM model from a task file; Review the ASM plan before calibration runs; Run the ASM workflow from the Codex project package · GitHub
On Co
OnCo is an open, cited knowledge graph of oncology, with a website, static JSON API, MCP server, CLI and a validated TypeScript corpus of cancers, treatments, trials and related records.
4 use cases: Let an AI agent search and compare On Co cancer records; Pull On Co cancer records into a script via CLI or API; Correct an oncology record and open a pull request · GitHub
Awesome Robot Use Agent
A curated GitHub list of papers, articles, open-source tools, benchmarks, and community demos for robot-use agents, which are AI systems that plan and act through robots.
4 use cases: Map the robot-use agent field from current surveys; Shortlist a robot-use method for a manipulation project; Choose benchmarks to test a robot-use agent · GitHub
Cambium
Cambium is a Python governance standard and toolset that keeps knowledge repositories maintained by LLM agents scoped, resumable, and under operator control.
3 use cases: Adopt Cambium governance for a knowledge repository; Resume an interrupted corpus task without guessing; Plan a long multi-batch corpus task with a Task Plan · GitHub
Open Jev
Open Jev is an untrained PyTorch research reconstruction of the ideas behind TypeSafe AI's Jev, which answers typed questions about nested state without generating text.
3 use cases: Run the bundled demo to see typed readouts in action; Prototype typed yes/no, choice, and score questions over JSON state; Test soft-target training with RLCDLoss on your own labels · GitHub
Moli
Moli is an open-source headless browser written in Rust that AI agents use to fetch pages, extract content, and automate browser tasks.
4 use cases: Fetch a web page as Markdown for an AI agent; Get a compact semantic tree of a page for a model; Capture a screenshot or PDF of a web page · GitHub
Effective Auto Research EAR
EAR is an open Python research workbench that helps with idea discovery, autonomous mathematical research, and evidence-linked reviewer responses.
4 use cases: Explore the EAR demo in a local browser studio; Turn a research direction into a ranked proposal; Inspect saved proof artifacts from the offline math run · GitHub
Awesome Rsi
Awesome RSI is a curated, community-maintained reading list of research on recursive self-improvement in AI, organized by model-level, harness-level, embodied, and safety topics.
4 use cases: Build a staged reading path on recursive self-improvement; Check RSI claims against the list's three evidence levels; Pick benchmarks for testing a self-improving agent · GitHub
Papergraph MCP
PaperGraph MCP is a Python tool that turns arXiv papers, local LaTeX projects, and born-digital PDFs into a local, evidence-grounded workspace of math results, proofs, and dependencies for AI agents and readers.
4 use cases: Map a math paper's main results before reading; Trace a theorem's dependencies before reading its proof; Find public full text for a DOI paper · GitHub
Modeling Workbench
A Codex and Claude Code workflow that guides math modeling contest work from official problem files to a checked, bilingual LaTeX and PDF paper, with evidence records for each step.
4 use cases: Check which modeling stages this machine can run; Start a contest project from official problem files; Test two candidate models before choosing one · GitHub
Misaka Agent
MISAKA is a Python multi-agent research system for the humanities and social sciences that turns one question into a reviewed research graph and a sourced research article.
3 use cases: Run a research question across several branches; Check each research claim against its source page; Expose the unstated premises in a draft argument · GitHub
Agentic STAR
AgentSTAR is a harness where a coding agent (Codex or Claude Code) reads a monocular video, builds an articulated object in Blender, poses it per frame, and iterates until the renders match the frames.
3 use cases: Reconstruct the garden shears example end to end; Turn your own video of an object into a 3D capture; Compare Codex and Claude Code on the same capture · GitHub
MemT 1 0
MemT-1.0 is an official research repository for a memory foundation model that keeps persistent, evolving memory inside the model's computation, served through SGLang with a Python client.
3 use cases: Run a chat that remembers facts across turns; Save a memory snapshot and resume it in a new session; Compare replies with native memory on and off · GitHub
Harness Zero
Harness-Zero is a Python research repo that distills an optimized agent harness into Qwen3.5-9B model weights, so the gains remain when the harness is removed at deployment.
4 use cases: Generate reviewed agent trajectories for training data; Distill a harness into Qwen3.5-9B weights; Adapt the harness bank to a new task domain · GitHub
Game the LLM Reviewer
An agent skill that packages editing strategies for making small, meaning-preserving wording changes to a finished paper before submission, aimed at known LLM reviewer preferences.
4 use cases: Run a meaning-preserving final pass on a finished paper; Tighten an abstract to 200 words without losing results; Tune a theory paper's stance, keep theorems and assumptions · GitHub
Jevry
Jevry is an MIT-licensed desktop browser agent (Electron, TypeScript) that carries out website tasks in plain language and reports results grounded in the visible page.
4 use cases: Compare two web pages and cite each claim; Find a library's getting-started guide in docs; Collect records that match filters in a web table · GitHub
Forest
FOREST is an open-source research agent framework that takes a research question through literature review, hypotheses, real experiments, analysis, and a full manuscript, keeping the work in an editable research graph.
4 use cases: Run a research project from question to draft paper; Run a research project from the terminal without the Web build; Clone a public GitHub repo and inspect its run diagnostics · GitHub
Agent Garten
AgentGarten is a Python repo with a neural renderer that turns depth and surface-normal data into video, plus training and streaming inference code; its code worlds and agent practice loop are not released yet.
4 use cases: Render comparison videos from a renderer checkpoint; Generate renderer video block by block with streaming; Train the renderer on your own depth and normal clips · GitHub
Code as World
Code-as-World is a Python research repo that provides local inference, QuantiPhy evaluation, OpenAI-compatible serving for its VL checkpoints, and a MuJoCo ballistic soccer example for physical reasoning.
3 use cases: Score Code as World checkpoints on QuantiPhy validation; Serve a Code as World model behind an OpenAI-style API; Run the bundled ballistic soccer MuJoCo simulation · GitHub
Harness Eval W
HarnessEval-W is a Python evaluation pipeline that scores generated world model videos with sub-agents and saves a traceable reasoning chain for each score.
4 use cases: Score a world model's generated videos with evidence trees; Check a finished run for missing scores before sharing; Run the bundled demo to test the setup end to end · GitHub
Jacobian
Jacobian is an MCP server, CLI, and Python library that gives AI agents a searchable set of typed, exact mathematical operations they can inspect and run.
3 use cases: Have an agent compute an exact matrix determinant; Compute a gcd and Bézout coefficients from the terminal; Let an agent find and run a graph or SAT operation · GitHub
Cadence
Cadence is a Python library for building simulated animal-like brains that keep learning from live experience, with an optional self-observing System 2.
3 use cases: Run a continuing brain that learns from live rewards; Preview a brain's responses without altering live memory; Test whether System 2 observers help a task · GitHub
Worldmodels101
A free interactive course on world models in AI, with nine chapters that each have an interactive view and a printable PDF.
5 use cases: Learn world models through nine interactive chapters; Generate printable PDF handouts for a chapter; Adapt the course for a new audience and keep the license · GitHub
Geo Sleuth
geo-sleuth is an agent skill made of Python scripts that estimates where a photo was taken, using OpenStreetMap geometry, elevation skylines, satellite imagery and street view, and reports its evidence.
3 use cases: Find where a photo without GPS was taken; Match a mountain skyline to a power line site; Pin a city street photo using public street-tree data · GitHub
Awesome Jev Gallery
A curated list of papers, open models, benchmarks, commentary and projects tied to TypeSafe's System One models and Jev.
4 use cases: Pick a local open model for typed decisions; Compare typed-decision models on shared benchmarks; Route Thai or Japanese intents with a small model · GitHub
Oc
A command line tool that turns websites into compact, numbered text views so AI agents can read and act on web pages with few tokens.
5 use cases: Check a package's version, license, and deps cheaply; Find a Python function's docs by exact name; Look up security advisories for a dependency · GitHub
Heretic
Heretic is a command-line tool that automatically removes safety alignment (censorship) from transformer language models using directional ablation tuned by Optuna.
4 use cases: Remove refusals from an open model in one run; Measure refusals and KL divergence against the original model; Decensor a larger model on a small GPU with 4-bit quantization · GitHub
Ksor
KSoR is an open-source TypeScript SDK for building a governed, authoritative knowledge record that humans and AI agents can operate from.
3 use cases: Stand up a governed knowledge site from a fresh project; Serve a governed record to AI agents with a refusal floor; Govern a team's policy or course content with human approval · GitHub
DSH Industry Research
A DeepSeek Harness domain pack for industry and company research that builds sourced chain maps, tracks public sources, writes company cards, and assembles auditable reports.
4 use cases: Map an industry chain with sourced values and gap list; Track policy and news for an industry in one timeline; Build company research cards from your own data files · GitHub
DSH Research Report
dsh-research-report is a DeepSeek Harness plugin that binds each report claim to immutable evidence snapshots, verifies those claims byte-for-byte, and seals the report with a recomputable manifest hash.
4 use cases: Seal a research report with byte-checked claims; Verify a sealed report from its directory alone; Find claims in a draft that lack source evidence · GitHub
Awesome Jev
A community-maintained, source-backed map of Jev (TypeSafe AI's System One decision model), linking official docs, integrations, papers, evaluations, and robotics notes.
4 use cases: Pick the Jev integration path that fits your stack; Route simple and hard tasks to different models; Hold risky agent tool calls for review · GitHub
Prysai LLM Playbook
A bounded, English-first LLM curriculum that teaches how to make one limited request to a language model, inspect the answer, and carry the method to a new task before choosing a platform.
4 use cases: Check an LLM answer before you act on it; Turn a vague request into an agent task card; Check a citation-shaped answer before you use it · GitHub
Vision HOPE
VisionHOPE is an official PyTorch implementation of a visual backbone built on self-referential nested learning, with ImageNet classification, COCO detection and segmentation, and ADE20K segmentation code plus standalone SRNL and operator modules.
4 use cases: Classify images with a pretrained Vision HOPE model; Add the Vision HOPE operator to your own vision network; Cut Vision HOPE inference time on an NVIDIA GPU · GitHub
Topconf Paper Figure Gallery
A searchable, filterable gallery of about 3,500 Figure 1 and teaser figures from six ML/AI conferences (2023 to 2026), plus FigureForge, a browser tool that drafts a Figure 1 from those references.
5 use cases: Find Figure 1 layout references for a paper; Draft a Figure 1 from your paper with FigureForge; Run the figure gallery on your own machine · GitHub
PAKGOV RAG Project
A research project that answers questions about the Constitution of Pakistan in English, Urdu, or both, using hybrid retrieval and citation-grounded answers, with a published retrieval evaluation.
3 use cases: Ask Constitution questions and get cited answers; Check the retrieval benchmark numbers yourself; Build a search index for Senate rules and acts · GitHub
Native Breeze TTS 2 Support for vLLM Omni
A learning project that adds Breeze TTS 2 speech synthesis support to the vLLM-Omni serving framework, with the official reference code and the modified framework source side by side.
5 use cases: Learn how a TTS model plugs into a serving framework; Run the CPU test suite on the Breeze integration; Diff the official Breeze code against the vLLM-Omni patch · GitHub
Simreal ML Bench
Simreal-MLBench is a public Python package with a 60-task agentic ML research benchmark, its scoring code, and a client SDK, while the official evaluation runs on a Simreal-operated service.
4 use cases: Validate the 60-task benchmark catalog locally; Recompute a published benchmark score from saved counts; Get an agent officially scored on the 60 tasks · GitHub
Knowledge Base Kit
Knowledge Base Kit is a Python toolkit that helps Claude Code or Codex turn documents into a local, source-backed Markdown knowledge base with full text, summaries, and a catalog.
4 use cases: Build a source-backed knowledge base from your documents; File documents into a knowledge base with Claude Code; Try the offline demo with no account or model calls · GitHub
Jev Social
Jev Social is an open-source, local-first research agent that runs read-only browser steps in your signed-in Chrome to produce source-cited reports from Instagram, TikTok, and LinkedIn.
4 use cases: Build a source-cited social research report; Let Codex run browser-grounded social research; Capture a TikTok video with its media as evidence · GitHub
Awesome REA Cases
A curated, auto-updated list of 627 reverse-engineering and AI-assisted reverse-engineering cases, projects, and ideas, with links to case pages.
3 use cases: Find a reverse-engineering case for an old device; Compare REA agent setups before a reverse-engineering project; Scope a firmware or driver fix from a documented case · GitHub
Thesys Core
Thesys is an open-source local research workspace for searching open-access papers, uploading PDFs, asking cited questions, comparing papers, and exporting citations.
4 use cases: Ask cited questions across your own PDF library; Find open-access papers and add them to your library; Check whether two papers agree on a specific claim · GitHub
Absent Author
An evidence list and two agent skills for spotting research papers that AI produced and no human stood behind, plus Python checks that flag candidate signs for a person to read.
4 use cases: Check an AI-assisted paper before you sign it; Grade a suspicious submission with evidence, not guesses; Scan a paper folder for AI-style surface traces · GitHub
Tren Torch
TrenTorch is a learning platform for machine learning, with a from-scratch NumPy framework built in a terminal CLI and a browser site with hundreds of coding questions.
3 use cases: Build a working ML framework from raw NumPy modules; Solve ML coding questions in the browser with instant grading; Run the web question checks locally before a pull request · GitHub
Second Brain OS
A starter kit and guide for building a knowledge base in plain markdown files that an AI agent builds, links and maintains for you, using Claude Code and Obsidian.
4 use cases: Turn saved articles into a self-linking wiki; Run a project through an Inputs, Process, Outputs pipeline; Audit an agent setup with the agents course plugin · GitHub
Jev Doc Search
A Python project that finds the page answering a question in a long PDF by searching a PageIndex tree with Jev's Choice and Noul primitives, without a vector database.
3 use cases: Find the page that answers a question in a long PDF; Locate the handbook page that answers an HR question; Ask many questions of one indexed annual report · GitHub
Awesome Jev Use Cases
A curated list of demos, repos, limits, cost figures and API notes for TypeSafe's Jev model, which answers typed questions instead of generating text.
3 use cases: Triage a large inbox with Jev for a few cents; Filter a social feed with a Jev classifier; Choose a browser agent demo before delegating a search task · GitHub
Diffusion OPSD
DiffusionOPSD is an external Python implementation of an on-policy self-distillation method for reward-guided post-training of diffusion models such as SD3.5-M and Z-Image-Turbo.
3 use cases: Post-train SD3.5-M toward a chosen image reward; Verify reward models load before a long training run; Train the Z-Image-Turbo backbone at 1024 resolution · GitHub
Veri Repro
VeriRepro is a Python package that checks whether a computational paper's claims are backed by its repository, environment, and execution evidence, and writes a PASS, FAIL, or PARTIAL reproduction report.
4 use cases: Plan a paper's reproduction without running its code; Verify a paper's metrics by running its code in a sandbox; Check paper discovery against the 15-paper smoke corpus · GitHub
Keystone Bench
Keystone is a Python benchmark that tests whether a clinical chat assistant changes its action when one decisive fact is removed, replaced, or contradicted, built on HealthBench conversations.
4 use cases: Audit a clinical chat assistant's response to missing evidence; Find which items a new assistant version fixed or broke; Confirm a local rebuild matches the published benchmark · GitHub
LLM Architecture From Scratch
A learning repo of PyTorch implementations of modern LLM architectures, with a table comparing each model's normalization, positional encoding, attention type, and use of Mixture of Experts.
3 use cases: Compare how current LLM architectures are built; Trace attention variants from code to table rows; Study Mixture of Experts layers in two models · GitHub
PairTeX
A local, Git-native review and feedback layer that renders an existing LaTeX repository to HTML so humans can comment and edit in place while a coding agent reads the recorded feedback.
5 use cases: Review a LaTeX paper in an HTML view and record feedback; Have an agent apply reviewer feedback to the TeX source; Try PairTeX on the bundled demo fixture · GitHub
Harness RSI
Harness-RSI is a companion survey for harness self-evolution, with a reading list, taxonomy data, a project website, and PDF downloads of the full report and short paper.
4 use cases: Choose which harness part to self-update first; Build a cited literature table for a harness project; Classify a new agent paper against the taxonomy · GitHub
AI Engineering Lab
A free, MIT licensed, self-paced 24-week AI engineering course built around one fictional freight company, with 43 runnable notebooks.
3 use cases: Start the 24-week AI engineering course today; Build a retrieval corpus from the freight case study; Learn coding agent harnesses with Claude Code or Cursor · GitHub
Evo VLM
EvoVLM is a research repo that uses LLM-guided evolutionary search to improve the inference programs for frozen Qwen3-VL-2B vision-language models on chart understanding.
4 use cases: Verify Evo VLM's recorded results on a CPU; Smoke-test exported Evo VLM programs without weights; Run the evolved Qwen3-VL-2B programs on your charts · GitHub
Other categories · All use cases · Hot repos
Questions
What is the best AI repo for Research in 2026?
Right now the top pick on Deduped is Scientific Agent Skills: A library of 177 Agent Skills that give AI agents documented, validated guidance for scientific packages, databases, and research workflows.
How are these Research repos ranked?
By how hot they are on GitHub (stars gained and mentions in the AI sources we track), then by how many practical use cases we found. The list is rebuilt every day.
Get 5 new use cases every morning. Free, one short email, unsubscribe anytime.
Get them free