The lecture you need, before you reach the parking garage.
Record an interesting case at the end of your shift. RobotParrot generates a tailored audio lecture — pearls, pitfalls, and evidence — ready to listen on the drive home.
A 2008 idea, finally possible.
Someday the best type of education will come from a little robotic parrot that sits on the shoulder of residents as they go and see patients. — Jeremy Joslin, MD · New England Journal of Medicine Horizons education-innovation challenge, October 2008
The vision was always the same: lectures that know what just happened on your shift. Fifteen years later, the parrot is real. Modern speech recognition, retrieval-augmented reasoning, and natural voice synthesis finally make it possible to turn a 90-second case summary into a tightly-targeted, evidence-grounded lecture in minutes.
Sign out your shift. Listen on the drive home.
Three steps, no typing, no reading. Built for the twelve minutes between your last note and your steering wheel.
Record one case
Tap the mic. Sign out a single patient's case like you would to a colleague — presentation, workup, decision, outcome. About 90 seconds. One patient per recording; if you've got three good cases, record three times.
We render the lecture
Modern AI models transcribe, parse the case, retrieve primary sources, draft a script, and proofread every dose, guideline claim, and statistic before voice synthesis.
Listen on the commute
A roughly eight-minute, second-person lecture referencing your specific case. Chaptered, with linked sources. Auto-pauses if you take a call. Resumes where you left off.
Under the hood — the full six-step pipeline
-
01
Talk through one case
Five guided prompts for a single patient: presentation, workup, management, what you weren't sure about, and how you think you did. About 90 seconds — one patient per recording.
-
02
Transcription
Speech recognition turns your audio into text. No PHI needed — speak as "the patient," not by name.
-
03
Case parsing
A language model extracts a structured case schema — diagnoses, vitals, interventions, and the uncertainty flags that drive the lecture's teaching agenda.
-
04
Evidence retrieval
PubMed and PubMed Central open-access full text, FDA drug labeling (DailyMed), and ClinicalTrials.gov — landmark trials, management controversies, and recent updates filtered to the last five years — plus a catalog of pre-validated teaching points maintained by our clinical educators.
-
05
Drafting & proofreading
One model writes the script in spoken prose; a second, independent model checks every dose, guideline claim, and statistic before it goes to voice synthesis. Pearls drawn from the educators' catalog are spoken as "our editors have taught that…".
-
06
Audio in your ears
A roughly eight-minute, second-person lecture — referencing your specific case — waiting on your phone before you reach the parking garage.
It opens with your case. It teaches to your uncertainty.
Your 58-year-old diabetic came in with crushing chest pain, diaphoresis, and ST elevations in II, III, and aVF — confirmed RCA occlusion on cath, DTB of 52 minutes, drug-eluting stent placed. You had two moments of hesitation, and both of them are worth unpacking completely.
First: the aspirin question. He told you he took some at home. You paused. Here's why you should never pause on that again… — Excerpt from a real Robot Parrot lecture (inferior STEMI case)
Right timing. Right teacher. Right venue.
Three things have to line up for didactic learning to actually stick. RobotParrot lines up all three.
Best timing for learning
When didactics can be immediately tied to experiential learning, retention is dramatically higher. The lecture lands while the case is still vivid — not next week, not next month.
Best person to learn from: you
When your learning is tied to the real case — the concerns, questions, and hesitations you flagged while telling it — the lecture is tailored to the specific, personal gaps that actually need filling.
Most convenient venue: your ride home
Reclaim the otherwise-downtime of the drive home. Learn while the case is still fresh — not just smart, but convenient. Use the time you reclaimed to get some extra sleep later.
Built to teach from the case, not the patient.
RobotParrot is a learning tool, not a medical record. It's designed so you never need to share a patient identifier — and it strips any that slip through before a transcript is stored or read by the lecture models.
You never need a name
The guided prompts ask for the clinical story — presentation, workup, decisions, outcome — not identity. Sign out the case the way you would to a colleague: speak as "the patient," never by name, MRN, or date of birth.
Identifiers stripped before storage
Every transcript runs through a two-pass de-identification first: pattern matching removes formatted identifiers — MRNs, dates, phone numbers — and an AI pass replaces any residual names or places with role descriptions ("a 54-year-old," "an outside hospital"). Only the de-identified version is saved or sent to the lecture models.
Audio doesn't linger
Your recording is used to make the lecture, then automatically deleted on a short retention schedule. What's kept afterward is the de-identified clinical case — not the patient. Always follow your own institution's policies on protected health information.
Pay per lecture. No subscription.
Tokens are sold in packs of 10. Each token covers one full lecture — transcription, research, drafting, proofreading, and voice synthesis. The price is set transparently from the underlying costs plus a flat creator fee.
Token pack
Good to know before your first case.
How long should a recording be?
Between 30 seconds and 6 minutes — aim for around 90 seconds. That's enough to sign out a focused case the way you would to a colleague. You'll get a gentle warning before the 6-minute cap so you can wrap up.
Can I record more than one patient at a time?
No — one patient per recording. RobotParrot builds a single lecture around a single case, so combining several patients in one take produces a muddled lecture (and still uses just one token). Got three interesting cases? Record three times — one lecture each.
What if RobotParrot can't make a lecture from my recording?
If a recording doesn't contain a real case — it's silent, it's a test, or it doesn't actually describe a patient encounter — RobotParrot won't generate a lecture. When that happens your token isn't spent: it's held while we take a quick look, then returned. You're only ever charged for a lecture you actually get.
What happens to my token if something fails on your end?
Any technical failure — an error while transcribing, researching, drafting, or generating the audio — automatically returns your token. You don't have to ask.
Do I need to avoid patient names and identifiers?
Yes — sign out the case as "the patient," never by name, MRN, or date of birth. The guided prompts never ask for identity, and even if something slips through we strip identifiers before any transcript is stored. See Patient privacy for how that works.