The user
Assessing clinicians
Running long, high-stakes sessions where attention belongs on the patient — not on wrangling documents and re-keying details between systems.
Case study Clinical · Observation tool
One place to capture the whole assessment.
Clinicians ran autism assessments across a tangle of documents and disconnected systems — slow, repetitive, easy to lose the thread. I designed a single structured workspace built so its notes flow straight into the report, and into AI summaries.
The core move — many scattered docs, one workspace
Word doc
Risk form
Consent PDF
Notes app
Report template
Email
One structured session
6+fragmented systems → a single source of truth
Before
After
01 Context
An assessment held together by copy-paste.
An autism assessment is hours of careful observation, mapped against dozens of clinical criteria, then written into a formal report. Clinicians were doing it across Word docs, separate forms and ad-hoc notes — capturing the same information twice, and stitching it together by hand afterwards.
The friction
Fragmented & repetitive
As one clinician put it: it's too hard to find and capture what you need across multiple systems — slow, complex, and easy to drop something that matters.
The brief
One structured space
Build a single in-session workspace that captures observations against clinical criteria — and feeds the report and AI summaries without manual copy-paste.
02 Approach
Designed around how clinicians actually work.
This was a 0→1 build with real weight — so it started with the people doing the job. I worked shoulder to shoulder with the clinical leads to map the assessment, then shaped the structure, the flows and the AI direction around it.
M1
Shadowing sessions
Sitting in on real assessments to see where clinicians lost time, switched tools, or re-entered the same information.
M2
Mapping the criteria
Working with clinical leads to translate the DSM-5 assessment structure into a workspace that mirrors how they think.
M3
Structured for AI
Defining how notes should be captured so they're clean enough to auto-populate a report and drive reliable AI summaries.
M4
Prototype & align
Bringing product, clinical and engineering to one shared vision, tested against live session scenarios.
01
Switching cost broke focus
Jumping between a Word doc, a risk form and a notes app pulled clinicians away from the patient and fractured the record.
So what
One workspace with everything — consent, risk, observations — in a single tabbed flow.
02
Notes weren't report-ready
Free-text scattered across tools had to be manually reshaped into the formal report — hours of work, and a place for errors to creep in.
So what
Observations are captured against clinical criteria, so the report half-writes itself.
03
Structure was the unlock for AI
You can't summarise a mess. Reliable AI summaries needed notes captured in a consistent, criteria-tagged shape from the start.
So what
I designed the capture model as the foundation the AI layer would stand on.
03 The design
Structure the chaos, then let it flow.
Three decisions turned a pile of documents into a workspace clinicians could trust — and a clean foundation for automation.
Decision 01
One session, tabbed by task
Consent, risk and observations live in a single session under clear tabs, autosaving as you go — so nothing is captured twice and nothing is lost.
One source of truth
Decision 02
Notes mapped to criteria
A criteria rail (A1, A2, B1…) lets clinicians file each observation exactly where it belongs, so the structure of the report is built as they work.
Report-ready by default
Decision 03
Built for the AI layer
Because notes are structured and criteria-tagged, they can be pulled straight into a draft report or an AI summary — clean input, trustworthy output.
Automation-ready foundation
Product The observation console
Everything in one calm workspace.
The in-session view: patient and autosave status always visible, the assessment split into clear tabs, and each observation filed against a clinical criterion in the rail. Dense information, made navigable.
Richard Hendricks ID: 024737374
Observation session
Autosaved just now
Consent & checks
Risk & MSE
Tasks & notes
Save & export
Misc
A1
A2
A3
B1
B2
B3
B4
Observation task
Headings
A1 · Social reciprocity
Conversation, shared enjoyment, emotional give-and-take, sequencing of interaction.
A2 · Non-verbal communication
Eye contact, gestures, facial expressions, body language.
B1 · Repetitive speech & movement
Stereotyped language, mannerisms, repetitive motor patterns.
B4 · Sensory processing
Responses to texture, sound, light, movement, temperature.
01
Context always visible
Patient identity and autosave status stay pinned, so a clinician never wonders whether their work is safe.
02
The criteria rail
A1–B4 down the side turns a blank page into a structured map of the assessment.
03
Progressive disclosure
Each criterion expands on demand, so the screen stays calm during a long session.
Product Consent & checks
Consent handled first, and clearly.
Recording and transcription need explicit consent before anything begins. The first tab makes that unmistakable — a plain-language prompt the clinician can read aloud, and clear options to capture what the patient agreed to.
Transcription & recording consent
Captured before the session begins
With your permission I'd like to record our session to help document our conversation. The recording is used only for your clinical care and viewed only by your clinical team. Are you happy for me to go ahead?
Consent to transcribeReceived
Consent to recordReceived
Consent not given
01
A script, not a guess
The prompt gives clinicians words to say, so consent is consistent and genuinely informed.
02
Explicit states
Transcribe, record, or not given — each is its own clear choice, never bundled or assumed.
03
Gated by design
Consent lives on the first tab because nothing else should happen until it's settled.
04 The payoff — AI summaries
Structured notes in, a trustworthy draft out.
This is why the structure mattered. Because every observation is captured against a criterion, the notes can be pulled straight into an AI-drafted summary — turning scattered session notes into the beginnings of a report. I shaped this as a draft for the clinician to review and own, never an answer to accept blindly.
Structured session notes
A1 · Social reciprocityLimited back-and-forth; responded to direct questions but didn't initiate or extend topics.
A2 · Non-verbalReduced eye contact during conversation; gestures used sparingly.
B1 · RepetitiveRepeated a preferred phrase at transitions between tasks.
B4 · SensoryCovered ears at sudden noise; sought firm pressure when unsettled.
AI-drafted summary
Across the session, reciprocal social interaction was reduced — the client responded to direct prompts but rarely initiated or sustained exchanges. Non-verbal communication was limited, with reduced eye contact and sparing gesture use. Repetitive language appeared at task transitions, and sensory sensitivities were noted around sudden noise.
A draft for the clinician to review & edit01
Clean input, clean output
The AI works because the notes are already criteria-tagged — structure is what makes the summary reliable.
02
A draft, never a verdict
The summary is a starting point the clinician reviews, edits and signs — the judgement stays human.
03
Hours saved, not corners cut
The tedious first-draft assembly is automated; the clinical thinking is protected.
05 Trade-offs
Where structure met the real world.
Structure vs. free flow
Rigid fields make clean data but can fight how a clinician naturally observes in a live session.
Structured⇄Fluid
Decision
A criteria rail to file against, plus a "Misc" space for anything that doesn't fit yet — structure without a straitjacket.
How far to trust AI
Summaries save hours, but an over-confident draft in a clinical setting is a real risk.
Automated⇄Accountable
Decision
AI drafts, the clinician decides — every summary is editable and owned by the person who signs it.
Density vs. calm
A full assessment is a lot of information; showing it all at once overwhelms a live session.
Complete⇄Calm
Decision
Tabs and progressive disclosure keep the whole assessment reachable but never all on screen at once.
06 Outcome
From scattered documents to one system.
One workspace, not six
Consent, risk and observation capture unified into a single autosaving session
Report-ready by design
Criteria-tagged notes flow into the report instead of being rebuilt by hand afterwards
An AI foundation, built in
Structured capture became the base layer that makes reliable AI summaries possible
Where it points
- Full AI-drafted reports from a completed session, clinician-reviewed.
- Live prompts that surface relevant criteria as an observation is typed.
- A shared assessment record across the clinical team, always current.
- Transcription that files spoken observations straight to the right criterion.
Rahman Malik — Product designer
Clinical · Observation & assessment tool