Independent Product · In Development
AI-assisted Rental Condition Report
I am building a clearer way for tenants and property owners to document a property’s condition at move-in and move-out. AI can suggest observations from photos. A person has to review them before they become part of the record.
DefinedExists now
- Problem, target users, inspection workflow, and MVP scope
- Data model: properties, rooms, inspections, photos, observations
- Mandatory human review before any AI suggestion is saved
ImplementingIn progress
- Inspection and photo-capture workflows
- AI-suggested observations from an uploaded photo
- Authentication and access control
PlannedStill ahead
- Private photo uploads as a published capability
- Remaining MVP features
- A public release
Not establishedNot claimed
- Users, adoption, or inspection volume
- Revenue or launch date
- Production AI accuracy
Inspection workflow
- 01Capture ImplementingRoom, item, and photo — inspection and photo-capture are being implemented
- 02Suggest ImplementingAI proposes an observation from the photo — it does not write the record
- 03Review ImplementingA person accepts, edits, or rejects the suggestion before anything is kept
- 04Record DefinedOnly reviewed text becomes part of the condition record
Data model
- PropertyThe place the record belongs to
- RoomWhere the inspection is scoped
- InspectionA move-in or move-out pass
- PhotoEvidence attached to that pass
- ObservationReviewed text, not a raw model output
AI suggestion → human review
- Photo ImplementingAn inspection photo is uploaded as part of capture.
- Suggestion ImplementingAI proposes an observation from that photo.
- Human review ImplementingA person accepts, edits, or rejects it.
- Approved record DefinedOnly then does the text become part of the condition record.
AI can suggest. A person decides what enters the record.
The user problem
The product is for tenants and property owners. At move-in and move-out they need a shared record of what the property looked like — not a later argument from memory.
I am building this independently. It is not a client assignment, and it is not released.
Why evidence is hard at the door
Move-in and move-out notes are often inconsistent, hard to compare, and too easy to dispute later. A photo dump in a chat thread is not a record. A paragraph written weeks afterwards is not one either.
The useful object is narrower than it first sounds: a structured condition record that both sides can look at, room by room, with the evidence attached to the observation.
What this product is not
The boundary is a condition record. It is not an automated damage judgment. It is not a legal outcome, and it does not claim a financial result from the record.
AI does not decide what happened to the property. If a suggestion is wrong, the human review is the point of the product, not a disclaimer added afterwards.
The inspection workflow
The defined flow is capture, suggest, review, record. Capture is room, item, and photo. Suggest is an AI-assisted observation. Review is a person accepting or editing. Record is only the reviewed text.
That sequence is designed. Inspection and photo-capture are being implemented. I do not have a public walkthrough of a completed inspection, so I am not showing one.
The data model
The record is modelled as properties, rooms, inspections, photos, and observations. A photo does not float on its own. An observation belongs to an inspection of a room in a property.
That model is defined. It is the shape I am building against, not a published schema or a live dataset.
Where AI fits
AI’s job is one step: suggest an observation from an uploaded photo. It can make the first pass faster. It cannot be the last pass.
I am not publishing accuracy claims. There is no production model performance I can stand behind, and I will not borrow one.
Why a person has to approve it
A condition record can follow someone into a dispute. Photos of a home are not a place to let a model speak unreviewed.
So the gate is mandatory: a human reviews and approves before a suggestion becomes part of the record. That is a product decision, not a later safety note.
What exists now
Discovery is done: the problem, the target users, the inspection workflow, the MVP scope, the requirements, and the phased milestones. The data model is defined. The human-review rule is decided.
What a visitor cannot do today is use a released product. There is no public app here, and I am not substituting a mock that looks like one.
What is still in development
In implementation: inspection and photo-capture workflows; AI-suggested observations from an uploaded photo; authentication and access control; the human-review step that sits in front of the record.
Still planned: remaining MVP features, and a public release. Private photo uploads are part of the product direction. Live-user validation has not started.
Privacy and access
The photos in this product would be of someone’s home. That is enough reason to treat them as private, even before anyone else uses the system.
Authentication and access control are part of the implementation. Private photo uploads are part of the product direction. I do not yet have a public, detailed security write-up — encryption choices, retention, or sharing rules — so I am not inventing one.
Current limitations
The product is not publicly released. There are no established users, adoption figures, inspection counts, or revenue. Production AI accuracy is not established. I have not published a completed inspection against real property data.
Those are limits, not temporary footnotes. They are the honest current state.
What I am learning by building it
I am shaping the product and learning the technical pieces needed to build it. The useful lesson so far is the boundary: define the record first, then give AI a narrow job inside it, then refuse to let the model write the thing that people will later trust.
How the implementation will change that judgment is still ahead of me. I will write that when there is a built path to point to, not before.
This page is a product decision record for work still in development. A later version can show a real, sanitized inspection path — only once that path exists, and only with data that is safe to publish.