YOUR RADIOLOGY REPORTING WORKSPACE
Your voice. Your report.
Dictate your findings. Generate a complete report in your own reporting style.
Browser dictation works now. Connect Groq with your personal API key for Whisper transcription and report generation, with no model download.
Tap a word to replay it. Highlighted words may need correction; unhighlighted words can also be wrong.
Review original recognized text
Only a few medical spellings are normalized. Review and correct the transcript before generating a report.YOUR DICTATION → YOUR COMPLETE REPORT
Speak freely. Report your way.
Connect Groq or your own server to generate reports without downloading a model.
Your reporting style 30 Oct 2026
Uses a snapshot of Dr Devkant Lakhera’s Dev Radiology Reporting skill and examination-specific references: acquisition beneath the title, detailed narrative findings, relevant normal lines, bold abnormal findings and a short dash-led impression. The model is instructed to preserve measurements, side and certainty. Tables are limited to supplied spinal AP canal diameters.
Generation uses your selected AI engine, separate from ChatGPT. Groq uses a compact version of your reporting rules and examination guidance to fit free usage limits. Your own server and on-computer AI use the full skill snapshot. Connected AI receives your report input; on-computer AI requires a model download. Review the draft against the examination. Skill changes in ChatGPT need a website update here.
Review the changes
Compare your reviewed dictation with the draft. These checks flag possible changes; they do not verify clinical accuracy.
Show wording changes
A FAMILIAR START
Your report structures.
Choose an examination, then add your findings and impression.
PICK UP WHERE YOU LEFT OFF
Your saved drafts.
Report copies saved in this browser, on this device.
LESS REPETITION, MORE FLOW
Words that sound like you.
Add a spoken phrase and the exact text you want it to become.
Add a shortcut
Use a distinctive trigger, such as “insert my normal liver”. Shortcuts expand during dictation and can contain a whole paragraph.
Shortcuts are saved on this device. They replace text after recognition; they do not train a medical speech model.
My shortcuts
GET COMFORTABLE
A workspace that listens.
Start with a short, anonymised test dictation.
Getting started
DevDictate has no website login. Browser dictation, editing and exports work without a developer account. Groq speech and report generation require your personal Groq API key; free-tier limits apply. Each visitor has their own report workspace.
- Choose an examination above the microphone. Add patient details only when appropriate for your selected service.
- For Whisper and report generation, click Set up free-tier AI, create a Groq key, and enter it in the connection window. Browser dictation can be used immediately without this step.
- Keep Rough dictation selected. Tap the microphone and allow microphone access.
- Dictate, then tap the microphone again to stop. Groq transcribes the complete recording after you stop.
- Check the transcript, then click Generate report. Your dictation stays available.
- Review findings and impression, then copy the report or export to Word, text or PDF.
Add to your phone
In Android Chrome, open this app’s link in a full browser tab. Use the browser menu → Add to Home screen. An embedded in-app browser may not provide microphone access.
Keyboard shortcut
Alt + Shift + D starts or stops dictation on desktop.
Your speech & your data
Groq (connected AI): your personal API key stays only in this open tab and is cleared on disconnect or reload. It is sent directly to the official Groq API over HTTPS, never saved in drafts or the website source. A connection check lists available models without sending audio or report text. Dictation sends the recording to Groq when you stop; Generate report sends the current report input and compact reporting guidance. Use a free-tier account to stay within its free limits; a key from a paid account can incur charges. Check your account limits and data controls. No other visitor receives your key. There is no automatic fallback to another provider.
My server: connect a server you control with its HTTPS address and access key. The key stays in this tab and is cleared when you disconnect or reload. Server speech sends your recordings to that address; server report generation sends your reviewed dictation and reporting style. The server package processes requests without retaining recordings or reports. Its hosting and models must be set up separately. There is no automatic fallback to a different service.
On-computer report AI (optional): report generation runs on your computer using WebLLM and Qwen3 4B. A compatible graphics processor with WebGPU and sufficient memory is required; it may be slow or unavailable on phones and older computers. The first use downloads roughly 2.5–3.5 GB from the model publisher and caches it when browser storage permits. Your dictation is not sent to a cloud AI for report generation. Browser speech is separate and may process audio online.
Whisper speech (optional): choose this engine only if you want offline transcription and its model download. It records locally and transcribes after you stop. Whisper Small English runs on the CPU without WebGPU; first use downloads a speech model of roughly 500 MB. Speech recognition is separate from report generation. The recording stays in this tab for playback until a new recording or case is opened. No recording is saved automatically. It is a general speech model and has not been validated for medical accuracy.
Browser speech (default): dictate live without downloading a speech model or installing an app. Your browser’s recognition provider may receive audio. Internet access may be required. This browser engine is separate from Connected AI and does not use your connected service.
On-device speech: experimental and available only on some browsers and languages. The app checks for a local language pack and offers a download when supported. It never falls back to online recognition automatically.
Reports: editing is local to the open tab. Connected AI receives the report input when you ask it to generate a report. Clicking Save draft stores a copy in this browser’s local storage. Drafts are not encrypted by the app or synced across devices. They belong to this browser and website address. Export your drafts before switching to another address or clearing browser data.
Clinical review: speech recognition and report generation have not been validated for clinical accuracy. Check negation, laterality, anatomy, names, numbers and units before using a report.
RIS/PACS: copy and paste your reviewed report into your reporting system. Direct typing into other apps is not included.
Voice command reference
Live navigation and “stop dictation” commands work with Browser live recognition. For Groq or offline recordings, tap the microphone to stop; commands are interpreted after transcription. The anatomical word “colon” and ordinary word “period” stay as text; say “colon mark” or “insert period” for punctuation. Hindi text works with the Hindi language setting; voice commands are in English.
If the microphone is quiet
Check the site’s microphone permission and your selected input device. Open the app in a full browser tab, keep it in the foreground, and try a headset in a quiet room.
If recognition is unavailable, you can still type, use templates and export. A “network error” means the browser’s speech service could not be reached; try another connection or a supported on-device language.
The animated bars indicate listening. Connected AI mode also shows a measured microphone level while recording.
Browser speech documentation ↗