AI · Applied Tools

Narrows symptoms through a structured question flow and ranks possible causes; CE-marked as a medical device. Unlike searching symptoms online it is disciplined: ordered questions instead of random pages, and causes ranked by likelihood.

Runs in the terminal and writes each change as its own git commit. Unlike editor-based assistants it gives you that safety net: revert a step you dislike with git revert.

Predicts structures of protein, DNA and ligand complexes with AlphaFold 3. Unlike ColabFold it needs no install or GPU and runs AlphaFold 3; in exchange, daily jobs are capped and commercial use is restricted.

Collects many documents into one library you can query together, with page-cited answers. That is where it parts from single-file tools.

Three lines gives a strong baseline on tabular data. Unlike training a single model, it trains several and combines them by stacking; frequently used as a Kaggle baseline.

A resource list for Actions on Google, the developer platform for Google Assistant. Since Google shut down Conversational Actions in 2023, unlike current assistant lists it now has only historical value.

A list of AI tools and projects that generate text, images and audio. Unlike single product pages it maps the field category by category and tracks its fast movement.

A list of methods for explainable AI (XAI) — showing why a model decided as it did. Unlike general ML lists it focuses only on interpretability.

Browser automation as a browser extension, with form filling, clicking and scraping steps assembled by drag and drop. Against Puppeteer or Playwright it runs inside your own session, so pages you are already logged into need no separate authentication.

A language model trained on PubMed abstracts, more accurate than general models at tasks like drug–disease relation extraction. Built for information extraction rather than generation. The repository is archived (last update July 2024); the model and code run as they are, but nobody maintains them.

Merges snippet search with chat. Unlike general chatbots, it searches repositories and docs and uses the hits as context.

Brings AI assistance to hardware and circuit design. Unlike software assistants it works outside the crowded field, on a thinly covered subject: electronics design.

Runs Node.js in the browser via WebContainers, so the generated full-stack app previews instantly without a server round trip. Unlike component generators such as v0 it runs the whole app, and errors are fixed in place.

Extracts data from websites and monitors them for changes without code — you train it by clicking in the browser. Against frameworks like Scrapy there are no selectors to write, and it attempts to self-heal when a page layout shifts.

Lives inside the scheduling tool, repackaging one post for each platform’s tone and length — removing the round trip to a separate copy tool.

A tool for asking questions of a PDF — upload and start without signing up. Unlike tools such as Humata it has the lowest friction; in exchange there is no library management or multi-document support.

An LLM agent that operates chemistry tools — RDKit, reaction databases, synthesis planners. Consulting real computation instead of the model’s recall sharply lowers error in chemistry.

A coding agent that creates files and runs commands inside VS Code, asking approval for each action. Unlike fully autonomous agents, that slowing step gives back the oversight you would otherwise lose.

A free, local tool that tracks token usage and cost across dozens of AI coding tools — Claude Code, Cursor, Codex, Gemini and more; run it with npx codeburn. Unlike providers’ billing dashboards, it breaks spend down by model, project and task.

An open code model and plugin with a Chinese interface and regional infrastructure — a working alternative where access to Western services is limited.

Google’s automatically generated documentation for codebases, useful for extracting structure and entry points the first time you open a repository. Unlike hand-written READMEs, it is regenerated as the code changes.

Speeds up sequence alignment with MMseqs2 to make AlphaFold runnable in Colab. Unlike AlphaFold Server, you can batch-process your own sequences without hitting a quota.

Summarises which way the literature leans on a question and shows the ratio of supporting to contradicting studies. Finding one paper and seeing the field’s tendency are different jobs.

An open-source coding assistant for VS Code and JetBrains, where you pick which model does completion and which does chat. Unlike GitHub Copilot the model choice is open — point it at a local model and code never leaves your machine.

A VS Code fork that indexes your codebase and makes multi-file editing the primary flow. Unlike a plugin it proposes changes as diffs and waits for approval.

A machine-learning library for drug discovery, including molecular featurisation and dataset loaders. Unlike general ML libraries, it takes on the tedious part of getting chemical data model-ready.

Rewrites the sentence and offers tone variants instead of flagging errors. Grammarly corrects, this rewrites — two different jobs.

Screens hundreds of papers and extracts fields such as sample size, method and finding into a table. Unlike general chat assistants, it is built for the screening step that eats the most time in a systematic review.

Simplifies whichever passage you highlight in a PDF. Unlike tools that summarise the whole document, it opens up only the one place you got stuck — which makes it narrow but accurate.

Open models and data pipelines for financial text, including fine-tunes for sentiment and news classification. An open research alternative to a Bloomberg terminal.

Extracts decisions, action items and questions rather than a raw transcript, and can write into a CRM. Otter concentrates on transcription; this on extraction.

In-editor completion and chat. Unlike its rivals it is strong on governance: in the enterprise plan, whether code is sent to a model and which models are allowed can be managed as policy.

Jupyter in the browser with a free T4 GPU. Unlike local Jupyter it needs no setup or hardware; in exchange sessions are time-limited and files vanish when the machine recycles, so long training runs need Drive mounted.

Estimates the probability that a text was AI-generated. Unlike plagiarism checkers it gives a probability rather than a matching source — its own docs acknowledge false positives, so the result should not be treated as proof.

English grammar and style checking through an extension that follows you everywhere you type. Unlike LanguageTool its style and tone suggestions are stronger; there is no Turkish support, so it only makes sense for English.

A distributed AutoML platform. Unlike library-only solutions, it also brings explainability and productionisation tooling, and can spill out of memory on large datasets.

Generates email, landing page and ad copy against a campaign goal, free and without requiring a HubSpot account.

A PDF assistant aimed at comparative reading of long technical documents. Unlike quick tools such as ChatPDF, each answer carries a source-page link so you can verify the claim in the document.

Asynchronous coding agent from Google Labs that connects to GitHub repositories to fix bugs, write tests, and bump dependencies in the background; opens pull requests directly rather than acting as an inline autocomplete.

Upload a CSV or Excel and analyse by chatting; it runs Python behind the scenes and produces charts. Unlike general chatbots it shows the code it used — which is what makes the result verifiable.

Notebooks sitting on the same platform as the datasets, with a weekly GPU quota more predictable than Colab’s. The real value is being able to read other people’s solutions.

A node-based visual analytics platform where the whole pipeline is saved and re-runnable. Unlike Python scripts it delivers reproducibility without writing code; common in enterprise data teams.

Rule-based checking for 30+ languages including Turkish. Unlike Grammarly it supports Turkish and can be self-hosted — a realistic option when the text must not go to a company.

Translates contract language into plain prose while preserving clause structure, so you do not lose which sentence came from which clause.

A self-hostable machine translation API running Argos Translate models. Against the Google Translate API the text never leaves your server and there is no per-request charge — lower quality, higher privacy.

Classic PDF chores — convert, compress, sign, OCR — with chat layered on. Broader than the single-purpose rivals and average at each individual job.

Builds an app from chat and takes on the Supabase wiring and deployment too. Unlike v0, which hands you a component, it aims at a running, database-backed product.

An open model fine-tuned for medical Q&A and evaluated on USMLE-style questions. For research and exam practice, not clinical use.

An open Gemma variant adapted to medical text and imaging. Unlike closed medical models its weights are open; it is for research — there is no regulatory clearance for clinical decision support.

An AutoML tool that automates model selection. Unlike other AutoML tools it emits a Markdown report with learning curves, importance plots and decision explanations — which matters when you have to present the result.

Concise prompt rules designed to eliminate generic, cliché language ("AI slop") from LLM outputs; provides direct negative constraints rather than lengthy style guides.

Answers clinical questions only from peer-reviewed literature, with citations, for verified health professionals. Unlike general chatbots it has source control — the very thing whose absence makes them unreliable in medicine.

An autonomous development agent with browser and terminal access, running in an isolated Docker container. Built to close an issue end to end, not to complete a line.

Visual-programming data mining whose widgets wrap scikit-learn. Lighter than KNIME with stronger visualisation, which is why it shows up in teaching.

A tool for asking questions of PDFs. Unlike PDF chat tools that require an upload, its browser extension works directly on PDFs you meet on the web — no download-then-upload step.

Stores snippets with their source and what you were doing at the time, so you can later search for “the thing I used while fixing that bug last week”. Unlike classic snippet managers it remembers context; it can run on a local model.

Decomposes a time series into trend, seasonality and holiday effects. Unlike classic models such as ARIMA it gives reasonable results without tuning; designed for analysts who are not statisticians.

compare_models() cross-validates dozens of algorithms in one line and tabulates the result, quickly narrowing which model family is worth pursuing.

Write a strategy and backtest it, with equity, options, futures and crypto data supplied. Unlike a home setup, the platform provides the costliest part — clean data; its LEAN engine is open source.

Produces alternative phrasings of the same sentence with adjustable length and formality. In academic writing it can raise plagiarism risk, so it needs careful use.

The foundational open cheminformatics library — SMILES parsing, substructure search, fingerprints. Unlike commercial cheminformatics suites it is open source, and most tooling in the field is built directly on it.

Start from one paper and expand the citation and co-author network visually. Unlike keyword search, it surfaces adjacent work you missed because you did not know its keyword.

Tailors a résumé to a specific posting and watches the formatting ATS parsers need. Unlike general writing assistants, it focuses on matching the résumé to the posting.

A coding agent forked from Cline. Unlike Cline, its architect, coder and debugger modes can carry separate system prompts and models — a strong model on planning, a cheap one on coding.

Reduces a paper to structured cards — core claim, method, limitations, references, each separate. Unlike tools that produce a flat summary it is structured — for the triage stage where you decide what to read in full.

Explains a paper paragraph by paragraph, unpacking a chosen equation or term in place. For a reader from outside the field, that inline explanation matters more than a summary.

A self-hosted completion engine written in Rust that fits on one GPU. The Copilot substitute when code must not leave the building.

A large directory of AI tools sorted by task, newest first. Unlike lists organised by company it starts from “what do I want to do?”; quality varies and many entries are thin wrappers.

Generates React components from text, using shadcn/ui and Tailwind so the output is copy-ready into an existing project. It produces pieces, not whole applications.

A Java toolbox of classical ML algorithms and one of the field’s oldest open projects. Behind modern libraries, yet still tidy for comparing algorithms side by side.

Cursor’s closest rival; its Cascade mode carries long tasks alone and runs terminal commands when needed. More autonomy, less oversight.

A natural-language service for developers that turns text and speech into intents and entities. Unlike services such as Dialogflow, it is owned by Meta and offered free.