Context is scattered.
A recording preserves the discussion, while the next step often lives somewhere else—in a note, a message or memory.
01 / SIMPLIFAI · RECALL / AI MOBILE APPLICATION
An AI conversation companion that transcribes discussions, summarises the meaning and suggests next steps—with the source and the final decision in your hands.
AI product concept · Human review · Mobile UX“Let’s aim for Friday, but confirm after the supplier responds.”
The team discussed Friday. Final timing depends on the supplier’s response.
Source · 18:42Overview
After a conversation, people still need to decide what was agreed, what remains uncertain and what they will do next. Recall focuses on that handover: from spoken context to a commitment the person has checked.
A recording preserves the discussion, while the next step often lives somewhere else—in a note, a message or memory.
“Let’s aim for Friday” should not silently become a confirmed Friday deadline.
Connect each suggestion to its source, make uncertainty readable, and ask for confirmation before saving an action.
How might we make conversations actionable while keeping people in control of their meaning?
Define the risky moment
Review patterns and assumptions
Map the information and journey
Move from wireframes to UI
Test comprehension and control
AI experience
Recall is designed around a reviewable AI workflow: turn a conversation into useful context without turning a model’s interpretation into an unchecked commitment.
Speech recognition produces a timestamped transcript after everyone agrees to recording.
A language model proposes a summary, decisions and possible actions. Conditional dates remain uncertain.
Ask retrieves relevant saved excerpts before composing an answer, with a link back to the original words.
The person checks the source, corrects the owner or date, and confirms what enters their task list.
“Let’s aim for Friday, but confirm after the supplier responds.”
Friday is tentative. Owner and due date need confirmation.
View source · 08:42Nothing becomes a saved commitment automatically.
Confirmed by youDesigned product behaviour, illustrated with sample content. This portfolio walkthrough does not record audio or call a live AI model.
Show “I couldn’t find that in your conversations” instead of a confident invented answer.
Keep the recording saved, explain the failure and offer a separate retry.
Keep the original words nearby. Let the person correct or dismiss the suggestion.
Research
I separated desk-research observations from questions that would require research with people. The design questions below guide the prototype; they are not findings from interviews conducted.
The opportunity is to make the transition between interpretation and action understandable on a small screen.
| Observed pattern | Question to investigate | Recall response |
|---|---|---|
| Record-to-task journey | When does a suggestion feel like a commitment? | A review step before an action joins the task list. |
| Multiple specialist perspectives | How do people interpret competing outputs? | One chosen lens, with the source always available. |
| Questions across saved conversations | What makes an answer trustworthy? | A source excerpt and a clear no-evidence state. |
| Consent and human-review guidance | Can people find control at the right moment? | A pre-capture check and per-conversation privacy controls. |
Interview guide / proposed
Recruit people who regularly turn meetings or catch-ups into follow-up work. Include people who take notes and people who rely on recordings.
No participants have been recruited or interviewed for this case study.Affinity map
This hypothesis board organises the review into four themes. The notes describe design concerns to investigate; they are not participant quotations or measured findings.
Knowing when recording begins
Pausing without losing the session
An alternative to audio capture
Design response: deliberate entry
Tentative dates read as deadlines
Decisions mixed with suggestions
Interpretations need a source
Design response: visible evidence
Actions need a clear owner
Review can happen later
Too many suggestions add work
Design response: confirmation queue
Who can access a conversation?
What happens when it is deleted?
Answers need boundaries
Design response: explicit controls
A project contributor moving between discussions, responsible for remembering decisions and coordinating the next step. The working need is continuity—not simply a longer transcript.
“When a discussion ends, help me recover the important context and confirm my next step, so I can follow through without turning every suggestion into a promise.”
Illustrative job statement, not a participant quote.User flow
The core journey asks permission before capture, preserves uncertainty during review, and requires confirmation before a task is created. Secondary paths support recovery and later recall.
Consent is explicit; a note is available as an alternative.
Pause and stop remain separate, visible actions.
Review the summary and open the source when needed.
Edit the owner or date, then save intentionally.
No agreement → Write a note
Interrupted analysis → Keep recording, retry
Unclear suggestion → Read source, edit or dismiss
Recent conversations
Reviews waiting
Start a capture
Saved conversations
Transcript and insights
Questions with sources
Confirmed next steps
Owners and dates
Open and completed
Wireframes
The first pass establishes six critical layouts. It gives capture one primary action, keeps the summary skimmable, and reserves a dedicated review screen for decisions that need attention.
Product catch-up · 2 suggestions need your review
Review suggestionsTell everyone that this conversation will be recorded and summarised.
Everyone has agreed to recording.
“Let’s aim for Friday, but confirm after the supplier responds.”
The team discussed Friday. Final timing depends on the supplier’s response.
Source · 18:42“Let’s aim for Friday, but confirm after the supplier responds.”
The date is conditional. Recall leaves it unconfirmed until you decide.
Faster, but it hides the distinction between an extracted suggestion and an agreed commitment. A single tap would make ambiguous dates too easy to accept.
One additional screen exposes the wording, owner, date and source. It adds a deliberate pause at the point where an interpretation becomes an action.
Visual system
Terracotta highlights the next action. Warm white surfaces and charcoal text keep the conversation readable.
Display 28 / body 16 / metadata 12–14
A consistent 4 / 8 / 16 / 24 spacing rhythm.
Icons support labels. Status uses words as well as colour. Capture controls are sized for touch, with clear focus states in the prototype.
App screens
Follow the app journey, or choose a step to explore.
Product catch-up · 2 suggestions need your review
Review suggestionsProduct catch-up · 2 suggestions need your review
Review suggestionsTell everyone that this conversation will be recorded and summarised.
Everyone has agreed to recording.
“Let’s aim for Friday, but confirm after the supplier responds.”
Preparing a summary and possible next steps.
The team discussed Friday. Final timing depends on the supplier’s response.
Source · 18:42“Let’s aim for Friday, but confirm after the supplier responds.”
The date is conditional. Recall leaves it unconfirmed until you decide.
Only actions you confirm appear here. Suggestions stay with their conversation.
Friday was proposed, but the team had not confirmed it.
The supplier’s response was the dependency.
Choose what to look for. The original conversation stays the same.
Possible actions, owners and open dates.
What was agreed, and what remains open.
Details to clarify before moving forward.
Choose what to keep and what to share for each conversation.
Record with everyone’s agreement, or begin with a written note.
Usability testing
The next step is a moderated study with a small mix of people who coordinate follow-up work. Five to six participants would provide an initial qualitative round, followed by iteration and a second round. No participant sessions or measured results are claimed here.
Ask participants to think aloud through a fictional team catch-up. Start without a feature tour. Observe the first attempt before offering prompts.
Capture wrong turns, assistance, interpretation of status, and the person’s explanation—not only whether they reached the last screen.
Include people who currently use manual notes, recordings, or AI summaries. Mix confidence levels with AI tools and include mobile accessibility needs.
Request permission for session recording and anonymise notes. Use sample conversations rather than personal or workplace recordings.
| Task | What to observe | Decision informed |
|---|---|---|
| Start a conversation without recording yet | Can the person tell when the microphone becomes active? | Consent wording and primary-action placement |
| Explain whether Friday is an agreed deadline | Does the person notice the conditional date? | Suggestion labels and uncertainty language |
| Find the words behind a suggestion | Can they locate and interpret the relevant source? | Source-link visibility and excerpt context |
| Correct an action and save it | Are owner, date and confirmation understood? | Review-screen hierarchy and form labels |
| Find an earlier decision with Ask | Do they inspect the reference before relying on the answer? | Answer structure and source prominence |
| Delete a conversation | Can they explain what will be removed? | Deletion wording and dependent-content warning |
Design decisions
The design review centres on three decisions to validate. These are prototype decisions, not changes proven by participant testing.
A conditional statement stays conditional. The user supplies the commitment.
A person can inspect the relevant sentence without searching the entire transcript.
The task list reflects deliberate choices, while unreviewed suggestions remain with the conversation.
Say that no supporting conversation was found. Suggest a narrower question rather than inventing an answer.
Keep the saved recording available. Show retry separately from starting a new recording.
Offer a written note. Explain microphone access only when the person chooses audio capture.
Business impact
Turn conversations into useful next steps without removing human judgment, creating value through better follow through rather than more generated content.
Less time spent rewriting notes into actions
Fewer missed decisions and unclear responsibilities
Stronger trust through editable and traceable suggestions
Measuring after release
These are success measures for post launch evaluation, not claimed results.
Value the time saved in review and follow up, then compare it with product delivery and model operating costs.
Final reflection
The concept keeps the original conversation close, shows what the system inferred and gives the person the final say before a suggestion becomes an action.