The report is no longer the center of your practice. Your patient is.
Introducing the PsychAssist.ai Summer Release
Get your day back and see more patients. Dr Barnes introduces the release that turns report writing from a burden into an outcome of a formulation you have already reached. Talk to your case, see where every claim came from, surface the signals in your data, and generate every supporting document from the finished report.
All built to a standard we derived from millions of data points and made public: what makes a report auditable, defensible, and true to the patient in front of you.
In this release
- Case chat
- Question the case in plain language; the report revises in place
- Source visibility
- Every generated claim resolves to the measure, document or observation behind it
- Derivative documents
- Every downstream document an evaluation owes, from the finished report
- Pipeline
- The full caseload by stage, on one screen
The problem
AI can now produce an assessment report that is fluent, well organized, and wrong. Often better written than a human first draft, and confidently incorrect in ways that only surface when someone reads it closely. Which, because it reads well, nobody does.
That is the problem. Good prose is what stops people checking.
An evaluation report cannot be assembled from template blocks. It has to be derived out of the entire case read in context: every raw test score, intake screener, session note and behavioral observation, weighed together. Anything working from a partial view of the case produces a plausible document.
And plausible is not defensible.
The expertise was never the bottleneck. The keyboard was.
For as long as assessment psychology has existed, the report has been the job. Not the interviews, not the testing, not the pattern only you can see - the eight hours afterwards, spent assembling and transcribing and formatting.
Caseload in this field has never been limited by how many patients a clinician can understand. It has been limited by how many reports they can type. Practices have quietly organized their hiring, their waitlists and their evenings around a typing constraint everyone mistook for a clinical one.
This release is built on the opposite premise. The report is not the work. It is the artifact of the work - and everything before it should carry the load.
We are the most expensive typists in healthcare.
Seven to ten years of training. A doctorate, a license, malpractice cover. And then the better part of a working day spent typing up a formulation we reached hours earlier - the profile, the rule-outs, the recommendations, all of it settled before a word goes on the page.
I built this because I was doing it myself every week, and I couldn’t accept that the limit on how many families I could see was how fast I could type.
What I didn’t expect was what the work would teach us. Once you have processed enough of it, the rules stop being opinions. They become the things that simply have to be true for a report to hold up - where every claim came from, what argues against your own conclusion, who approved it, whether it reads as you. So we wrote them down, and we published them.
What’s in this release
Four capabilities, and one that has been quietly proving the point for months.
Agentic case chat
Take a meeting with your case.
Ask the case a question in plain language and get an answer drawn from that patient's own data. Why this diagnosis held. What contradicts it. How this presentation differs from a similar one you saw last month. Where a score sits against the rest of the profile.
Then ask for the change - and the report revises in place, in your voice, with the sourcing intact.
Or start from where most clinicians start.
- Defensibility check
- Peer review this draft
- Second opinion on the diagnosis
- Signals in the noise
Before: you scrolled between documents to reconstruct your own reasoning, then retyped the paragraph.
Cite and source
Every sentence knows where it came from.
Every generated statement carries its source on the sentence: which measure, which uploaded document, which session note, which behavioral observation you recorded at the time.
Not in a support ticket. Not on request. On the sentence, while you are reading it.
Before: you took the sentence on trust, or you went and checked it yourself.
Convergent and divergent evidence
The case against your own conclusion, surfaced.
Every diagnosis the system surfaces is tested against that patient's own dataset and presented three ways: what supports it, what converges on it from another direction, and what diverges from it.
Differential diagnosis has always required weighing the evidence that argues against you. This is the first system that puts that evidence in front of you by default rather than leaving you to remember to look for it.
Every bullet is cited to its source. You decide what carries into the report.
Before: the disconfirming evidence was in the data. Whether it reached the report depended on how tired you were.
Approval gate
Nothing reaches the record unapproved.
Uploaded protocols and score reports are read into structured facts, each one carrying the informant it came from and how confident the extraction is.
You see the source document beside the extraction, correct anything wrong, and approve. Only then does it become available to the report.
Before: extracted data went straight into the draft, and you found the errors later, in the finished report.
Secondary documents
One assessment. Every document it owes.
An evaluation does not end at the report. There is the school report, the summary for the referring physician, the letter to the therapist, the plain-language guide the family actually reads.
Generate all of them from the finished report - from a template you control or a plain-language instruction - each in its own editor, each exportable.
Before: four more documents, written from scratch, in the evenings.
Pipeline
Your whole caseload, one screen.
Every patient, every stage, at a glance - from referral through testing to signed report.
This one shipped earlier in the year and it is the reason we are confident about the rest. It was the first time we treated the assessment as a system rather than a series of documents. Everything above is what that made possible.
The patterns you would have found on your third read
Across every measure in the case, the system surfaces cross-instrument formulations - the constellation rather than the individual scores. Anxiety-driven perfectionism presenting as a compensatory strategy. An executive functioning profile that explains an achievement gap the achievement testing alone would not account for.
Each one is presented with its central, supporting and diverging evidence, cited. You choose what carries into the report and what does not.
This is what a block editor cannot do. Assembling paragraphs is a formatting problem. Seeing a pattern across nine instruments is a reasoning problem.
We didn’t set out to write a standard. The work wrote it.
Every capability above exists because of a problem we hit building this - and the problems turned out to be the same ones every system in this category has to solve, whether or not it admits to them.
You cannot generate a defensible report from chunks of plausible text. You have to know where every claim came from. You have to surface what argues against the conclusion. You have to gate what enters the record. It has to read as the clinician who signs it.
Those are not preferences. Across millions of data points they are the things that separate a report that reads well from a report that holds up.
So we wrote them down, published them, and hold ourselves to them in public.
- Provenance - Every clinical claim resolves to an identifiable source.
- Disconfirmation - The evidence that argues against a conclusion is surfaced, not only the evidence that supports it.
- Approval - Nothing machine-generated enters a record without a logged clinician approval.
- Voice - The report reads in your voice and follows your standards, because you are the one who signs it.
- Auditability - Every action carries a timestamp, an actor and a role.
- Disclosure - AI use is stated in the record itself, not buried in a license agreement.
AI proposes. The clinician decides. That is the whole architecture, and it is the reason the rest of this release is possible.
The report is an outcome of a well-crafted workflow.
That is what we have been saying in every demo for two years. This release is the first time the platform fully delivers on it.