Construction Field Assistant working title

The problem

A construction project generates enormous context: plans and revisions, photos, texts, change orders, schedules, estimates, punch lists, and the informal knowledge in a foreman’s head. Almost none of it is reachable from where the work happens. A superintendent standing on a slab with gloves on cannot search a shared drive, and the answer they need is split across three apps and a text thread from two weeks ago.

The result is familiar to anyone in the trades: time lost driving context around — calling the office, scrolling for the right photo, rebuilding information that already exists.

The research question

What does an assistant need to know — and how must it listen and speak — to be useful on an active job site? Field conditions are hostile to every default assumption of AI products: screens are hard to use, audio is noisy, connectivity is unreliable, and nobody has attention to spare for a chatbot.

Why existing approaches are insufficient

Generic chat assistants assume a quiet room, a keyboard, and a user willing to type context into the model. Construction software assumes an office. Neither meets a person mid-task whose question is short, whose hands are busy, and whose answer lives in a specific drawing revision or a specific customer file. Retrieval here is not a semantic-similarity problem — it is an operational one: the right document for this job, this phase, this crew, today.

The product

A voice-first assistant, carried in a pocket, that answers from the company’s own project context: plans, photos, schedules, estimates, contacts, and notes. Ask what changed in the latest kitchen revision, capture a field note against the right job, or pull the customer’s selections — without stopping work to operate a screen.

Development status

This project is in research and product development. We are actively investigating field workflows and building the retrieval and voice foundations; the capability list above is the design target, not a shipped feature list. Nothing on it should be read as implemented today. We are looking for contractors and construction companies interested in shaping the pilot — start a conversation if that could be you.

What this project demonstrates

This project shows how DataKnife approaches an industry from the workflow side first: field research before feature lists, retrieval design driven by how job-site information actually lives, and interface decisions (voice-first, glanceable, interruption-tolerant) derived from physical working conditions rather than from what is easy to build.