Replace Suki on infrastructure LifeMD already owns. Same provider experience. Same EHR integration. Audio never leaves our perimeter.
Three structural reasons to evaluate a self-hosted alternative — and the integration shipped to dev only two months ago, so the switching cost is at its lowest.
Suki bills monthly per provider. As LifeMD scales the provider base, this cost scales linearly with no leverage on unit economics.
Every encounter audio + transcript transits Suki's cloud. That's the largest third-party PHI surface we currently maintain.
The 23-section LOINC schema and Elation pipeline are now coupled to one vendor's roadmap, prompts, and pricing decisions.
Audio in → byte-compatible Suki JSON out. Live on Spark today — Parakeet is transcribing, the local LLM is generating notes in Suki's exact 23-section schema. The existing thecvlb/lifemd → Elation pipeline doesn't change at all.
Stereo recording with provider on the left channel, patient on the right.
PR #15133 endpoint
createNonVisitNote
Real output from this Spark — not screenshots. Pick a sample, pick a model, click Run live to watch the note stream out token-by-token, or jump to the Mic tab and dictate a note while you're talking.
Click start, speak as the patient. Parakeet streams your audio to ASR. Server-side voice activity detection ends each turn. The doctor (MedGemma 27B + Piper TTS, all on Spark) replies out loud and asks follow-up questions. End the consultation and we generate the structured note from the whole conversation.
Suki output from production (Justin's March 11 thread) vs. ours from OpenBioLLM-70B on Spark. 21 fixed sections + dynamic problem sections. LOINC codes hardcoded by our backend; the LLM never picks codes.
The patient presents for a consultation regarding weight management.
- Obesity - Hypertension
- Lab tests before medication for weight loss - Continue nutrition and physical activity - Wegovy: side effects can include nausea, stomach discomfort, constipation - Follow-up after lab results
- Height: 5 feet 8 inches - Weight: 215 pounds
The patient presents for a consultation regarding weight management.
- Obesity - Hypertension
- Lab tests are needed before starting weight-loss medication - Maintain current nutrition and physical activity - Wegovy: be aware of nausea, stomach discomfort, or constipation early - Plan follow-up once labs return
- Height: 5'8" - Weight: 215 lb
And we own the prompt, the schema, the model — swap to whatever beats this on the day it ships.
Click a phase for details. The proof shipped during the hackathon — phases 2-4 are the engineering investment we're asking leadership to greenlight.
Spark + Parakeet ASR + 3-model A/B → Suki-shape JSON → live Elation write for one test patient.
Wrap into svc-transcribe with Presidio redaction, Celery async, ECS deploy, monitoring + alerting, audit logs.
NVIDIA Riva for production gRPC streaming, Sortformer diarization, Triton multi-tenancy.
Real-time partial transcripts during the visit, voice commands, multilingual via Canary-1B.