Roadmap

Two hundred needs, twenty per family. These are not entries: no content is produced for them until somebody writes them. They say where the catalogue is going, and what there is to pick up.

25 published 0 drafts 175 to write 200 au total

How to contribute

Detect and filter 20

  • Block abusive login attempts Spot waves of password guessing against a login form without punishing forgetful users. Take it on
  • Blur faces in photos Hide the faces of the people caught in an image before publishing or sharing it. Take it on
  • Detect text copied between documents Find passages reused from one document in another, even when a few words were changed. Take it on
  • Filter bot traffic out of server logs Separate human visits from automated crawls so that traffic figures mean something. Take it on
  • Filter noisy alerts before paging someone Group and drop repetitive or harmless alerts so on-call is woken only for a real incident. Take it on
  • Find credentials left in source code Find the access keys, passwords and tokens left in a repository before they get published. Take it on
  • Find duplicate records in a customer database Find the records that describe the same person despite different spellings, abbreviations or missing fields. Published
  • Find personal data in a spreadsheet Know which columns of a file hold personal data before sharing or archiving it. Take it on
  • Find the same image twice in a media library Find visuals that are identical, or merely cropped, resized or recompressed, cluttering a library. Take it on
  • Mask personal details in a message Replace the phone numbers, emails and addresses in a message before displaying, storing or forwarding it. Published
  • Moderate public comments Keep insulting or hateful comments off a public page without silencing legitimate criticism. Published
  • Screen the files users upload Check that an uploaded file really is what it claims to be and carries nothing harmful. Take it on
  • Spot a fraudulent payment attempt Flag suspicious orders at checkout without turning away legitimate buyers. Take it on
  • Spot explicit images among user uploads Keep sexually explicit uploads from ever appearing on a public site. Take it on
  • Spot fake reviews Flag mass-produced or paid-for reviews before they distort the rating shown for a product. Take it on
  • Spot profanity in usernames Reject crude usernames, including the ones disguised with digits or lookalike characters. Take it on
  • Spot scam links in messages Catch phishing or shady redirect links posted by users before a reader clicks one. Take it on
  • Spot spam in a contact form Keep automated and promotional submissions out of a public form without blocking genuine enquiries. Published
  • Tell whether a text was machine-written Estimate whether a submitted text was written by a language model, and how much that estimate is worth. Take it on
  • Turn away disposable email addresses Refuse throwaway addresses at sign-up, since the account becomes unreachable within hours. Take it on

Extract 20

  • Collect the contact details listed on a page Gather the phones, emails and addresses shown on a web page to fill in a record. Take it on
  • Extract amounts and currencies from a text Find the sums quoted in a message or a document, with their currency, however they are written. Take it on
  • Extract dates from free text Find the dates mentioned in a message or a document, whatever their format. Published
  • Extract measurements from a product description Find the dimensions, weights and volumes quoted in free text and bring them to a common unit. Take it on
  • Extract opening hours from a business page Turn opening hours written in plain words into slots a program can compare against the clock. Take it on
  • Extract the details held in a CV Recover the contact details, jobs, studies and skills from an application received as a file. Take it on
  • Extract the fields of an invoice Pull the number, date, total and supplier out of an invoice, whatever its layout. Published
  • Extract the ingredients and steps of a recipe Separate the ingredient list, with its quantities, from the sequence of preparation steps. Take it on
  • Extract the key clauses of a contract Locate the term, the deadlines, the notice period and the termination conditions inside a contract. Take it on
  • Extract the key terms of a document Surface the terms that characterise a document, so it can be indexed, labelled or previewed. Take it on
  • Extract the tables from a PDF Recover, as rows and columns, the tables inside a document laid out for print. Take it on
  • Find the people and organisations named in an article List the proper names mentioned in a text and tell what kind of entity each one is. Take it on
  • Isolate the newest reply in an email thread Separate the message that just arrived from the quoted history and signature wrapped around it. Take it on
  • Pull product data from a shop page Read the name, price, availability and images of a product shown on a shop site. Take it on
  • Pull quotes from an interview transcript Find, in a transcript, the passages each speaker said that stand up as quotes. Take it on
  • Pull the main content out of a web page Recover the article text without the menus, banners and side panels around it. Take it on
  • Split a bibliography into references Split a reference list and recover the author, title, year and source of each entry. Take it on
  • Split a postal address into fields Separate street, postcode, town and country from an address typed on a single line. Published
  • Split log lines into fields Turn heterogeneous log lines into named fields so they can be filtered and counted. Take it on
  • Spot a tracking number in an email Find the parcel reference in a shipping confirmation and tell which carrier it belongs to. Take it on

Classify and route 20

  • Assign expenses to accounting categories Give each expense line the accounting category it belongs to, from its label and its merchant. Take it on
  • Classify a company by its line of business Assign a company to a sector using its name, its description or its website. Take it on
  • Detect the language of a text Work out which language a message is written in before processing it or passing it to the right person. Published
  • Dispatch jobs to the right processing queue Send each incoming job to the queue that suits its size and urgency, to avoid pile-ups. Take it on
  • Gauge the tone of a customer review Tell whether a review is positive, negative or mixed, to follow overall mood and react to the worst ones. Take it on
  • Group free-text survey answers into themes Surface the recurring subjects in a batch of open answers instead of reading them one by one. Take it on
  • Identify the intent behind a user message Work out what the person wants — cancel, ask, complain — so the right next step can start. Take it on
  • Label audio clips by what can be heard Tell an alarm, a voice and background noise apart in a recording, so the right reaction follows. Take it on
  • Label contract clauses by type Find which clauses in a contract cover confidentiality, termination or liability. Take it on
  • Label log lines by severity Tell what deserves an alert from what is only noise, in a stream of heterogeneous logs. Take it on
  • Map free text to an official code list Match a hand-written description to the exact code of a mandated, very long classification. Take it on
  • Put products in the right catalogue category Place each item in the catalogue tree using its name and its description. Take it on
  • Recognise the type of an uploaded document Tell an invoice from a contract or a receipt as soon as it lands, so it enters the right process. Take it on
  • Route a request to the smallest model that fits Decide up front whether a request needs a rule, a small model or a large one, before handling it. Take it on
  • Route support tickets to the right team Send every incoming request to the team able to handle it, without someone reading the whole queue first. Published
  • Sort incoming email into folders File incoming messages into useful folders instead of letting a single inbox overflow. Take it on
  • Sort photos by what they show Organise a batch of images by their content, so a given scene or object can be found later. Take it on
  • Spot urgent messages in a shared inbox Bring the messages that cannot wait to the top, without reviewing the whole inbox by hand. Take it on
  • Tag articles by topic Put one or more topic labels on articles so they can be grouped and found again later. Published
  • Tell whether a message needs a human reply Separate what an automatic reply can settle from what needs a person, without letting a sensitive case slip through. Take it on
  • Add a search page to a small website Offer site search on a modest website without standing up a dedicated infrastructure. Take it on
  • Add autocomplete to a search bar Offer useful suggestions from the first letters typed, fast enough to keep up with the typing. Published
  • Answer questions from an internal knowledge base Get a written, sourced answer drawn from in-house documentation instead of a list of links to open. Take it on
  • Combine filters and keywords in a product search Let a visitor narrow down by size, colour or brand while still typing free text. Take it on
  • Find visually similar images Find the pictures in a library that look like a given image, without going through keywords. Take it on
  • Find where a quoted passage comes from Identify the source document of a copied extract, even when it was slightly reworded or cut short. Take it on
  • Forgive typos in a search box Return useful results even when the query holds a misspelling, swapped letters or a missing accent. Take it on
  • Handle synonyms and in-house jargon Match the word a user types with the term the documents use, when the two are not the same. Take it on
  • Keep a search index in step with changing data Make a change show up in the results without reindexing everything or waiting until the next day. Take it on
  • Make a pile of scanned documents searchable Search for a word inside documents that exist only as scanned images. Take it on
  • Match a postal address to a known place Find the official address behind an approximate, incomplete or badly formatted entry. Take it on
  • Match misspelled company names Recognise that the same company appears under several spellings across two files of different origins. Published
  • Query a database in plain language Let someone who does not write queries pull a figure or a list out of a database. Take it on
  • Search a large archive of logs Find the trace of an incident in logs piled up over months, without losing the day to it. Take it on
  • Search across past conversations Track down what was said in chat threads or meeting notes. Take it on
  • Search in one language, find results in another Query a document collection written in several languages from one language and still get the right passages. Take it on
  • Search in your own documents Find a piece of information across personal or company files without remembering which file holds it. Published
  • Search inside a single long document Find the useful passage in a manual, a contract or a report of several hundred pages. Take it on
  • Search several sources from one search box Query separate tools at once and present the results as a single, coherent list. Take it on
  • Search through a codebase Find where a function is defined or used in a project too large to read end to end. Take it on

Recommend 20

  • Blend editorial picks with automatic suggestions Let a single list hold both the newsroom's hand-picked choices and computed results. Take it on
  • Choose which comments to show first Order the replies under a post so the most useful ones are read first. Take it on
  • Explain why an item was suggested Give each suggestion a plain reason, drawn from the data that produced it. Take it on
  • Find visually similar images Offer, inside a picture library, the images that look like the one on screen. Take it on
  • Give fresh content a lift Give recent posts a chance without pushing the established favourites out of the ranking. Take it on
  • Keep suggestions from looking alike Remove near-duplicates from a list of suggestions so it covers several interests. Take it on
  • Match offers with requests Pair two lists on a marketplace: the listings published and the needs people posted. Take it on
  • Offer a replacement for an out-of-stock item Find the closest item still available when the one the customer wanted is missing. Take it on
  • Personalise a home page feed Assemble each signed-in visitor's home page from what they have looked at before. Take it on
  • Rank products by relevance Order the products on a catalogue page so the most relevant ones come first. Published
  • Recommend a clothing size Advise which size to order from the item's measurements and past purchases kept or returned. Take it on
  • Recommend the next video Pick what to play next when a video ends, based on what was just watched. Take it on
  • Recommend the plan that fits the usage Tell a customer which offer matches their actual usage rather than what they declared. Take it on
  • Suggest a reviewer for a change Point to the person who best knows the files touched by a change awaiting review. Take it on
  • Suggest accounts to follow Offer a member other accounts on the network, starting from the ones they already follow. Take it on
  • Suggest products that go together Offer, next to a viewed item, what customers usually buy along with it. Take it on
  • Suggest recipes from what is on hand Find dishes that can be made with the ingredients already there, allowing a few gaps. Take it on
  • Suggest similar articles Show, at the foot of an article, other pieces on the site about the same subject. Published
  • Suggest something to a brand-new visitor Offer a first sensible item when no history at all is available about the visitor. Take it on
  • Suggest the next action in an app Highlight the step the person is likely to take next, from the journey so far. Take it on

Predict 20

  • Announce a waiting time Tell people in a queue how long they still have to wait, and hold to it. Take it on
  • Anticipate a stock shortage Warn before an item runs out, combining how fast it leaves and how long restocking takes. Take it on
  • Check that a model is not drifting Watch forecast quality over time and raise a flag when the model comes loose. Take it on
  • Estimate a delivery's carbon footprint Work out the emissions tied to a shipment from the transport mode and the distance. Take it on
  • Estimate a second-hand price Offer the seller a range consistent with comparable items already sold on the site. Take it on
  • Estimate a text's reading time Show the reader how long an article will take before they start it. Take it on
  • Estimate attendance at an event Work out how many people will turn up, so welcome desks and supplies can be sized. Take it on
  • Estimate how long a job will take Give a likely duration for a planned job, based on comparable jobs already carried out. Take it on
  • Estimate the no-show rate for a slot Work out what share of a slot's bookings will go unused, so filling can be adjusted. Take it on
  • Fill a gap in a series of readings Rebuild the missing values of an interrupted record without distorting the averages drawn from it. Take it on
  • Find the seasonality in a series Bring out the weekly or yearly cycles in a measurement before trying to forecast it. Take it on
  • Forecast site load Estimate the traffic expected so the infrastructure is sized before a known peak. Take it on
  • Forecast staffing needed for a shift Estimate how many people must be on hand each opening hour, from past activity. Take it on
  • Forecast the coming weeks' cash Project the money coming in and going out to see whether the balance will hold. Take it on
  • Forecast weekly sales Estimate the coming weeks' sales from past figures, in order to plan purchasing. Published
  • Give a delivery date Tell the customer when their parcel will arrive, from delays actually seen on that route. Take it on
  • Plan maintenance before the breakdown Read a machine's measurements for the signs that precede a failure, and step in on time. Take it on
  • Put an uncertainty range on a forecast Pair a forecast with an honest range, so the reader knows how far to trust it. Take it on
  • Spot a lasting change of trend Tell a real turning point from a passing wobble in a curve watched every day. Take it on
  • Spot an anomaly in metrics Flag when a tracked measurement leaves its usual behaviour, without drowning the team in alerts. Published

Generate 20

  • Build a colour palette from an image Pick the colours of a page from the photo or the logo it contains. Take it on
  • Build preview images for shared links Compose the image that goes with a page when its link is shared elsewhere, without drawing it by hand. Take it on
  • Build quiz questions from a lesson Create test questions on a teaching text, with wrong answers that remain plausible. Take it on
  • Draft replies to common emails Offer a ready-to-review answer for the requests that keep arriving in a shared inbox. Take it on
  • Draw a diagram from a description Get a diagram of a process or an architecture described in a few sentences, and be able to correct it afterwards. Take it on
  • Give every account a default avatar Show a recognisable thumbnail for the people who have uploaded no picture. Take it on
  • Personalise a mass mailing Adapt each message of a campaign to the recipient's data without writing a thousand letters by hand. Take it on
  • Produce client code from an API schema Get the calling functions for a service from its published description, without writing them one by one. Take it on
  • Produce invoices as PDF files Produce a printable document from the lines of an order, with a consistent layout. Take it on
  • Produce placeholder images Show a stand-in image wherever the real one is still missing, in a mock-up or an unfinished catalogue. Published
  • Produce test data sets Fill a database or a mock-up with believable data, without ever pouring real customer records into it. Published
  • Turn a question into a database query Let a non-technical person query data by asking the question in plain language. Take it on
  • Turn titles into readable web addresses Turn a page title into a short, stable, readable address, despite accents and repeated titles. Take it on
  • Write a monthly report from figures Turn a table of results into a written commentary, regular and comparable month to month. Take it on
  • Write alt text for images Describe every image on a site for the people who cannot see it. Take it on
  • Write error messages people can act on Tell the person what went wrong and what they can do about it, without technical jargon. Take it on
  • Write page descriptions for search engines Give each page a short description sentence, different from the ones on the site's other pages. Take it on
  • Write product descriptions Write the sales text for every item in a catalogue, starting from its characteristics. Published
  • Write release notes from the commit history Say what changed since the previous version, in terms that matter to the people using it. Take it on
  • Write several variants of a message to test Produce several wordings of the same pitch to see which one works best with the audience. Take it on

Transform 20

  • Apply a house style to a text Enforce the in-house writing rules: agreed vocabulary, capitals, wordings that are ruled out. Take it on
  • Clean up a messy CSV file Make a carelessly exported file usable: shifted columns, dubious separators, headers that keep changing. Published
  • Expand the abbreviations in a text Replace acronyms and shorthand with their full form, for reading aloud or for search. Take it on
  • Fix the spelling and grammar of a text Find and fix the mistakes in a text before publication, without changing its meaning. Take it on
  • Flatten nested data into a table Turn an API response into rows and columns, ready for a spreadsheet or a database load. Take it on
  • Get the text out of a PDF Pull the readable content out of a document, whether it already carries text or only a page image. Take it on
  • Map fields between two formats Feed one system with another's data when the field names and the values do not line up. Take it on
  • Put phone numbers into a single format Unify numbers typed in twenty different ways so they can be compared and dialled. Take it on
  • Reduce a web page to its content Drop menus, banners and adverts to keep only the article, its headings and its paragraphs. Take it on
  • Remove the background from a photo Cut a product or a person out of the scene to show it on a plain background. Take it on
  • Repair text with a broken encoding Restore the accents and characters mangled by a bad hop from one system to another. Take it on
  • Resize and crop images for the web Produce a photo in the display sizes you need without cutting people's heads off. Take it on
  • Rewrite a text in plain language Make an administrative or legal notice readable for someone outside the trade. Take it on
  • Shorten a text to fit the space available Bring a heading or a summary down to the room available without making it unreadable. Take it on
  • Split a long text into coherent passages Prepare a document to be indexed or handled piece by piece, without cutting through an idea. Take it on
  • Straighten and clean up a scanned page Make a skewed, grey or spotted scan usable before reading it or filing it away. Take it on
  • Strip quotes and signatures from emails Keep only the message actually written, without the quoted history or the footer notices. Take it on
  • Summarise a long document Get the essentials of a report or a contract without having to read all of it. Published
  • Translate interface strings Make an application exist in another language, keeping the placeholders and the room available on screen. Published
  • Update translations after the source text changes Spot what moved in the source version and retranslate only that. Take it on

Recognise and transcribe 20

  • Align a transcript with its audio Match each word of an existing text to the moment it is spoken in the recording. Take it on
  • Check that a scan is usable Spot straight away that an uploaded image is blurry, cropped or too dark, and ask for another. Take it on
  • Detect the language of an audio recording Find out which language an audio file is spoken in, to send it to the right processing. Take it on
  • Find the spoken segments of a recording Split a long recording into speech and silence segments before transcribing or archiving it. Take it on
  • Get the text out of a screenshot Pull the text shown in a screenshot sent by a user, to analyse or search it. Take it on
  • Identify a music track from a short clip Find which musical recording matches a short audio excerpt, noise included. Take it on
  • Know whether a PDF needs OCR Tell a PDF that already carries its text from one made of images, before running a costly pass. Take it on
  • Read a barcode from a photo Get the reference encoded in a barcode photographed with a phone, even skewed or badly lit. Take it on
  • Read a mathematical formula from an image Convert a printed or handwritten formula into notation reusable in a document. Take it on
  • Read a QR code in a camera stream Detect and decode a QR code live in the camera image, without asking for a photo. Take it on
  • Read a utility meter from a photo Read the index shown on a water, gas or electricity meter photographed by a technician or a customer. Take it on
  • Read handwritten notes Turn handwritten notes into text, crossings-out and uneven handwriting included. Take it on
  • Read text from a scanned page Turn a scanned paper document into text that can be searched, copied or processed. Published
  • Read the ticked boxes on a paper form Find out which boxes were ticked on a printed then scanned questionnaire, without retyping it. Take it on
  • Recognise a spoken command Trigger an action from a small set of spoken phrases, on a device or inside an application. Take it on
  • Recover a table from a scanned document Recover the rows and columns of a printed table and rebuild a usable file. Take it on
  • Spot a specific sound in a recording Detect an expected sound, an alarm, an impact or a struggling machine, in a continuous audio stream. Take it on
  • Straighten a crooked scan Detect the skew and orientation of a scanned page and set it straight before any reading. Take it on
  • Subtitle a video Produce a synchronised subtitle file from the soundtrack of a video. Take it on
  • Transcribe a meeting recording Get the text of what was said in a recorded meeting, to read back or search. Take it on

Decide and validate 20

  • Apply discount rules to a cart Decide which discounts apply to an order, in what order, and which ones do not stack. Take it on
  • Check a company identification number Check that a company identifier is well formed and matches a genuinely registered company. Take it on
  • Check a configuration before deployment Spot a missing, inconsistent or dangerous value in a configuration file before it reaches production. Take it on
  • Check a file is really the format claimed Make sure an uploaded file really is the format its extension promises, before processing it. Take it on
  • Check a password against a policy Reject a weak or already-leaked password without turning sign-up into an obstacle course. Take it on
  • Check a product catalogue for inconsistencies Find mixed units, contradictory categories and empty fields in a catalogue before publishing it. Take it on
  • Check an invoice against its purchase order Check that an incoming invoice matches what was ordered and then delivered, line by line. Take it on
  • Check bank details before a transfer Catch a typing error in an account identifier before committing a payment. Take it on
  • Check that a contract contains the required clauses Check that a contract document carries every required clause before sending it for signature. Take it on
  • Check that a postal address is deliverable Make sure a typed address exists and can receive a parcel, before starting a shipment. Take it on
  • Check that a translation file is complete Find missing keys, forgotten variables and untranslated strings before shipping a new language. Take it on
  • Decide when to retry a failed operation Choose whether a failed operation should be retried, set aside or reported, depending on the error. Take it on
  • Decide whether a change needs human review Send for human review only the changes that deserve it, and let the rest through automatically. Take it on
  • Detect conflicts in a schedule Find overlaps, double-booked resources and impossible slots in a schedule. Take it on
  • Reconcile a bank statement with a ledger Match bank operations with accounting entries and isolate the ones that do not line up. Take it on
  • Validate a form server-side Check that the data received from a form is complete and coherent before storing it. Published
  • Validate a phone number and its country code Check that a typed number is valid in the stated country and bring it to a single form. Take it on
  • Validate an API payload against its contract Cleanly reject an incoming request that breaks the published contract, with a usable message. Take it on
  • Validate an email address without sending a message Work out whether a typed address stands a chance of existing before sending anything to it. Take it on
  • Validate an imported file against its expected schema Check column by column that a customer-supplied file matches the agreed format, and say what is wrong. Take it on