Dictate your patient notes, flawlessly.
VoiceChart.io is a B2B SaaS platform serving independent physician practices and mid-sized hospital groups. It captures ambient clinical conversation via browser-based audio streaming, transcribes and semantically structures the encounter in near-real-time, and delivers a formatted, review-ready clinical note in under 90 seconds. The platform integrates with Epic and Athenahealth via FHIR R4 APIs, keeping existing EHR workflows intact while eliminating the documentation burden that drives physician burnout.
The client's pilot survey of 40 primary care physicians revealed a brutal reality: doctors were spending an average of 3.1 hours per day on after-hours charting — colloquially called "pajama time." Existing voice-to-text tools produced raw transcripts that still required heavy manual editing. Structured templates were rigid and slow. The documentation gap was directly correlated with physician attrition, patient throughput limits, and compliance risk from incomplete notes submitted days after the encounter.
The engineering challenge was equally sharp: streaming audio, low-latency transcription, and clinical NLP had to work together reliably across variable network conditions in real clinical environments.
I designed and built a streaming pipeline that captures microphone input in the browser, chunks audio every 250ms via WebSockets, and feeds it to a self-hosted Whisper instance for continuous transcription. A fine-tuned GPT-4 prompt layer then classifies utterances into SOAP note sections in real time, surfacing a live structured draft on screen as the encounter unfolds. Physicians review, make minor edits, and sign — the entire post-encounter workflow collapses to under two minutes.
The architecture was built for fault tolerance: Redis-backed session state ensures a dropped connection never loses a partial note, and all audio is discarded after processing to satisfy HIPAA requirements.
I served as sole architect and lead engineer across the full stack. Responsibilities included:
VoiceChart.io ships a focused, high-impact feature set built around the actual clinical workflow:
After a 90-day rollout across two pilot practices covering 18 physicians and approximately 4,200 encounters per month, the numbers were unambiguous. After-hours charting dropped 58%, with the average note completion time falling from 8.4 minutes to 1.9 minutes. Physician satisfaction scores on internal surveys rose from 5.8 to 8.7 out of 10. Note completeness scores — measured against the practice's existing quality rubric — improved by 22%, largely because the ambient model captured clinical details physicians previously omitted when typing under time pressure.
The platform processed over 12,000 encounters in the pilot period with 99.94% uptime and zero HIPAA incidents. One practice reported a measurable increase in daily patient throughput after physicians reclaimed time previously lost to documentation.
Tell our assistant what you have in mind — it'll sketch the first version of your game plan on the spot, and we'll pick it up from there. No forms, no waiting.