Studio
A small collective. The voice on this site is we, because the work is staffed that way.
Everydayintelligence was founded in 2026 in Singapore, on Bukit Pasoh Road, five minutes from Outram Park. We are a collective studio: the same group that sits in discovery still reads the evaluation misses after launch. We do not publish a team page, and we do not put named portraits on the work, because the engagement is held by the studio rather than by a rotating account lead. In our first year we have kept the bench small on purpose, so that a client can reach the people who actually wrote the prompts.
The room is a working floor. Client sessions happen at the table. Prototype weeks happen at the desks along the window. When a corpus cannot leave the client’s building, we work there instead, and the Bukit Pasoh room becomes the place we write the notes at the end of the day.
Organisation
One owner on your side. One studio contact. A weekly written rhythm.
Each piece of work has a single studio contact who writes the weekly note and holds the open questions. Retrieval design, extraction, assistant behaviour, and evaluation sit in the same conversation. We do not hand a prototype from a research group to a delivery group. The person who collected the evaluation questions is the person who scores them in month three.
You name one counterpart who can walk us onto the desk and who can decide, in a week, whether a miss is acceptable. If that person cannot be named, we wait. Shared drives, model keys, and repositories live in your tenancy from the first working week. We keep working copies only as needed to build, then we delete them when the handover is signed.
No chat box goes to a whole department until an evaluation set has been scored at least once in the open. The set is made of questions staff already typed. A fluent miss is recorded as a miss. We would rather keep a prototype with ten people for another fortnight than decorate a launch that we cannot yet measure.
Retrieval systems cite a file and a page, or they say they cannot find a source. Assistants inherit the same rule for anything that looks like a fact from the corpus. Invention is treated as a product defect. The refusal text is written with the client, because a night-desk operator will act on whatever appears on the screen.
We do not bind the work to a single vendor. The evaluation set is the quality contract. When a cheaper or faster model appears, we rerun the set and write down what would be lost. The client decides. Hosted models and open weights can sit in the same engagement if the corpus is split by residency.
Runbooks, scoring scripts, and a named owner on your side are part of the product. If care continues month to month, it is a continuation of those files, kept in your tenancy. If you want us to step away entirely, we will, after the first monthly score has been read together.
Inside a week
Monday we send a short written note: what moved, what failed, what we need from you. The note is the meeting. A call happens if a decision cannot be made in writing. Mid-week we are on the desk or in the files, pairing with the people who do the work. Friday we rerun whatever slice of the evaluation set is in play and put the misses at the top of the next Monday note.
There is little theatre. You will see unfinished interfaces, red scores, and questions we have not yet decided how to treat. You will also see the refusal cases, which are often the most useful hour of the week. If a model upgrade is tempting, we run the set before anyone flips a switch, even when the upgrade is cheaper.
Visitors to Bukit Pasoh find a small floor, a table, and a map of the current evaluation questions on the wall during prototype weeks. We are here Monday to Friday, 09:00 to 18:00 SGT, and closed on weekends and public holidays. If the corpus cannot leave your building, the week looks the same, only the table moves.
Who we sit with
Thirty to eight hundred people, one painful desk, a counterpart who can decide.
We work with operations, professional services, trade, and claims teams that already have documents, a roster, and a system of record. The company is usually between about thirty and about eight hundred people. Smaller than that, the overhead of an evaluation set rarely pays. Larger than that, we still can, if the first problem is owned by a single desk.
We are a poor fit for a brief that wants a public chatbot, a slide about transformation with no named operator, or a system whose job is to replace a licensed sign-off. We are a better fit when someone on your side is already losing hours to search, retyping, or the same midnight questions, and is willing to sit with us while we measure that.
Visit
Bukit Pasoh Road, five minutes from Outram Park.
Write with a short brief, or telephone during studio hours. We reply within one working day and will tell you whether a discovery sprint is the right next step.