Features — full tour

Every piece of brtlb, in the order you'll meet it. Skim the headers; dive into whichever section is relevant.

Recording a visit

The Home screen has one giant graphite button: Record visit. Tap it. Mic permission prompts (once). brtlb is recording.

That's the whole flow. No new-visit form, no patient picker, no template selection up front. By default brtlb is in ambient mode — captures the whole room with speaker separation. If you'd rather narrate the note yourself, tap the small or dictate instead link below the Record button for classic dictation (no diarization).

While you're recording

The recording UI is deliberately subtle in ambient mode — designed for the patient sitting across from you to barely notice it:

• A small pulsing red dot with status text ("Ambient · Recording") — universal "recording" indicator at calm size.
• A modest timer counting elapsed time.
• A breathing line below the timer that gently widens with audio level — conveys "mic is alive" without VU-meter drama.
• For the first 4 seconds after start, a full 24-bar meter appears so you can confirm the mic is picking up audio, then it auto-collapses. Tap Mic check any time to bring it back.
• Dictation mode keeps the full meter on — no audience to perform for.

Action buttons

• Pause / Resume — when the visit hits a pause (parent steps out, you go examine an ear). Pauses don't count against transcription time.
• Mark moment — tap any time you want brtlb to pay extra attention to what's happening right now. The button flashes seafoam; later, when the note is generated, brtlb is told "the physician marked these moments as important." Useful for nuanced findings, a worried-parent moment, the part where the kid said something quotable. No pop-up, no label entry — single discrete tap. Mark as many as you want.
• Stop — when you're done. brtlb saves the audio and immediately starts the pipeline.

After Stop

You land on the Review screen. The pipeline runs automatically:

uploading → transcribing → generating → ready

A small status banner at the top tells you which stage you're in. Total time is usually under 30 seconds for a 15-minute visit. Long visits take proportionally longer; brtlb waits up to 90 minutes for AssemblyAI to finish transcribing — covers any realistic visit (most 60-min recordings finish in 3–5 min of processing, 90-min autism evals in 5–8 min).

The Review screen

Two big panels side by side (or stacked on mobile):

Tweak with AI

The hero of the note screen. A textarea that takes plain English:

"Shorten the assessment"
"Rewrite the plan as a numbered list"
"Add return precautions for fever"
"Patient is 8mo, not 8yo — fix throughout"
"Include that mom is a peds nurse and we discussed
 home Tylenol dosing"

Type your instruction. ⌘+Enter (Mac) or Ctrl+Enter (Windows) to send, or click Revise note. brtlb returns the COMPLETE revised note — not a diff. The transcript still gates fabrication, so changes are grounded in what was actually said.

This is how most physicians will edit. Faster than direct markdown editing, and you don't need to remember any conventions. Just tell it what's off.

Speaker chips & roles

Above the transcript, brtlb shows a chip for each detected speaker label (A, B, C, …). Tap any chip to assign a role: Parent, Patient, Provider, Sibling, or Other. The transcript and any future regenerate use those labels.

brtlb auto-detects roles on the first pass — provider vs. parent vs. patient — so most visits don't need manual chip tagging. When auto-detection misfires (a kid not speaking up much, an unusual role distribution), the chips are the fastest fix.

Multiple patients in one recording

brtlb handles back-to-back encounters in a single ambient recording. If you record two well-checks in a row, or a sick visit then a sibling's quick check, brtlb keeps them separate automatically.

After transcription, brtlb runs a split-by-patient pass that reads the diarized transcript and identifies which utterances belong to which child. Patient labels come from names actually mentioned in the visit ("Tommy", "Lily") with an ordinal fallback ("Patient 1") if names don't surface.

For each patient, brtlb generates a separate note with the visit type the splitter detected and only that patient's relevant utterances. A patient tab strip on the Review screen lets you view, copy, or tweak each note independently. Use the All combined tab for one-shot copy of everything.

The split is purely transcript-based — brtlb has no schedule access, no EHR connection, no list of who's coming today. It just listens to who's being talked to and about.

Mixed visits (well-child + acute complaint on the same kid) are preserved correctly — the splitter keeps it as a single segment with visit_type: well_child, includes_preventive_care: true, and the acute concern in the segment metadata.

Auto template detection

brtlb watches the transcript and picks the best template without you having to. The nine built-ins:

Ambient mode auto-routes after transcription. The picked template shows up in the Template dropdown on the right panel. Don't like the pick? Change the dropdown and click Regenerate — brtlb re-runs the LLM with your chosen template (no re-transcription).

Want your own? Custom templates →

Adaptive note length

brtlb sizes the note to the visit. A 5-minute URI gets a focused note; a 60-minute mental-health follow-up or 90-minute autism eval gets a longer, richer one. No template "default length" override — length is a function of clinical content density.

You can always steer manually with Tweak with AI: "shorten the assessment," "expand the plan with return precautions."

Visit chapters (long visits)

For ambient recordings ≥30 min, brtlb generates a small chapter map above the transcript: 3–7 named segments with timestamps and a one-sentence summary. Examples:

0:00   Parent interview — 6-month history of social withdrawal
12:35  Child observation — play, communication, response to name
32:10  Discussion of findings

Lets you scan a 90-minute autism eval transcript at a glance instead of scrolling. The chapter detector runs automatically when the recording qualifies; short visits skip it.

Review warnings — the safety net

Below the note, the Review warnings panel runs an independent LLM pass that hunts for two failure modes:

• Hallucination — note says something the transcript doesn't support.
• Omission — transcript discusses something the note misses.

Plus three secondary checks: mixed-visit collapse (well-child + acute reduced to one), assessment/plan mismatch, and wrong-patient risk (sibling contamination, name drift).

The reviewer is told the transcript may have STT errors and to interpret charitably — the goal is flagging real safety issues, not nitpicking phonetic transcription mistakes.

Output is a small markdown bullet list with severity icons:

• 🔴 Critical — safety-relevant
• 🟡 Warning — clinically meaningful but not unsafe
• ⚪ Info — minor

Click Check for warnings when you want a second look. Re-run after edits to see if you've cleaned up the issues.

Sensitive content flag

The Review warnings panel runs an extra check: if the transcript touches suicidality, self-harm, abuse, substance use, sexual activity, eating-disorder behaviors, custody conflicts, or intimate partner violence — brtlb adds a 🟡 (sensitive content) line naming the topic.

Doesn't redact, doesn't block sharing. Just a heads-up to review before pasting to a shared chart so adolescent or sensitive disclosures don't accidentally end up in a parent-visible note.

Quotes captured

Below the note, the Quotes captured panel pulls up to 5 verbatim parent/patient quotes from the transcript on demand. Useful for:

• HPI ("My son has been saying 'I don't want to be alive' for two weeks").
• Safety screens — verbatim language matters medicolegally.
• Pediatric phrasing that's clinically meaningful in its specificity.
• Cases where what was said exactly is the documentation.

Strict guardrails: no paraphrase, attribution required (Parent / Patient / Sibling), STT-garbled quotes are dropped rather than guessed. Returns "No quotes captured." when nothing qualifies.

Clinical pearls — the collegial layer

A separate on-demand pass that surfaces 0–3 short clinical observations about this visit — patterns, subtle differentials, family dynamics worth noting. Like a senior colleague leaning over after you wrap up.

Examples:

Pearls are NOT generic safety advice and NOT restatements of your plan. If brtlb doesn't have anything genuinely useful to add, it returns "No pearls."

Sharing the note

Three section-paste modes plus four global export buttons.

Section paste modes

Export buttons

• Share — native share sheet (AirDrop on Apple, Android Share). Falls back to copy on desktop.
• Copy — copies the whole note with bold formatting preserved when the destination accepts rich text (Elation, Word, most rich-text fields). Plain-text fallback for everything else.
• Email… — opens your mail app with the visit label as subject and note as body. Only safe if your email provider is HIPAA-BAA-covered.
• Download — saves a .txt file with the visit label as the filename.

There's no "post to EHR" button. brtlb is BYO-EHR — you copy or share into wherever your charts live.

Privacy controls

Settings → Privacy & Security has the levers that matter:

• Audio purge. 7 / 30 / 90 days or "never." Old audio blobs auto-delete after the window; transcript and note metadata stay.
• AssemblyAI auto-delete on completion (default on). brtlb fires DELETE /v2/transcript/{id} seconds after the transcript returns. Toggle off only if your practice policy needs longer vendor-side retention.
• Idle lock. Lock the UI after N minutes of inactivity. Ongoing recordings keep recording behind the lock screen — silence auto-stop is paused while locked so a hidden banner can't kill your visit.
• Raw transcript JSON export — diagnostic link on every Review screen header. Downloads the stored AAI output for offline review (no PHI leaves your device).

Want the full design story? Why brtlb works this way →