Otto v1: why we built an AI report writer for chartered surveyors

In the summer of 2025 we shipped the first working version of Otto, an AI-assisted report writer built for chartered surveyors completing RICS Level 2 HomeBuyer Reports. This post is an honest account of why we built it and what we learned in the first months.

The problem we kept hearing about

We talk regularly with surveyors who list on or partner with Home. The workflow complaint came up often: inspecting a property takes a few hours, but writing the report can take the same again, sometimes longer. The inspection is the skilled part. The writing is necessary, but the time spent on it is not proportional to the clinical judgement involved.

A surveyor completing several inspections a week can spend a large part of their working hours in front of a blank document, reconstructing observations from handwritten notes and photographs. That is time that cannot be billed differently. It is also time that accumulates into real pressure on capacity.

We thought we could help with that specific part of the workflow, without touching the judgement part.

What v1 did

Otto v1 had two main capabilities.

The first was audio dictation. During or immediately after an inspection, the surveyor speaks their observations into Otto using a phone or laptop. Whisper, OpenAI's speech-to-text model, converts the audio to a structured transcript. The surveyor can dictate room by room, element by element, in their natural order and at their own pace. We go into how that pipeline works in how Otto transcribes a property inspection dictation.

The second was image analysis. The surveyor uploads photographs taken during the inspection. GPT-4o analyses each image and identifies visible defects or conditions relevant to RICS element categories. Those observations are attached to the relevant section of the report, a process covered in more detail in how Otto reads condition from a property photograph.

From those two inputs, Otto generated a draft report in the RICS Level 2 structure. The surveyor then reviewed, edited, and signed off the final document. Otto did not sign anything. Otto did not decide anything. Every condition rating in the published report was set or confirmed by the surveyor.

What we deliberately did not do

We made a firm decision not to let Otto assign condition ratings autonomously. A condition rating in a RICS report is a professional judgement with legal weight. It affects whether a buyer proceeds, what they pay, and what they can reasonably claim if something goes wrong later.

We built what we called a verification step. Before any report was completed, every AI-suggested element was checked against the surveyor's own stated rating. If they disagreed, the surveyor's rating won, always. This was not a fallback feature. It was a core design constraint from the start, and it is the same constraint we later formalised and wrote about in the rating fidelity problem.

Where v1 shipped

The first version of Otto was built inside the homedata app between July and August 2025. In October 2025 we moved it into the Atlas monorepo, which powers the wider Home platform, so it could share infrastructure and be developed alongside our other surveyor tools.

What came next

The move to Atlas enabled us to build faster. Over the following months we extended Otto to cover Level 3 Building Surveys alongside Level 2 (see RICS Level 2 vs Level 3: which survey do you need? for the difference), improved the dictation pipeline, and began work on the writing-style personalisation that became one of the more interesting technical problems we solved.

Otto v2, which launched in June 2026, brought extended context handling and an auto-mode that reduces the manual steps between inspection and draft. We will write more about that in a later post.

For now, you can read more about what Otto offers on the surveyors tools page, or learn about what we build for surveyors more broadly at RICS surveyors on Home.

Further reading: RICS on home surveys.