YC S24
Real-time AI Engineering at SimCare
Conversations,at human speed.
First hire at SimCare AI. Building the real-time systems behind AI client conversations for healthcare training.
Behind the experience
Explore the flowStart with the person.
Speech becomes text that the system can work with.
I worked with Deepgram transcription and async backend workflows to connect incoming speech to the conversation pipeline.
Keep the conversation moving.
Language-model inference produces the response.
The FastAPI backend orchestrates streaming inference as part of a live conversation, with latency considered across the entire request.
Close the loop naturally.
Generated speech returns to the live video experience.
I connected the speech and language layers with the product surfaces and infrastructure that support face-to-face AI interaction.
An illustrated overview of the work. Select a stage to look closer.
Listen
01 / The challenge
What had to
feel simple.
SimCare lets healthcare learners practice with AI clients and get feedback. Those conversations need to move at a human pace. Transcription, language models, and speech generation each take time; connecting them means paying attention to the whole experience, from the first spoken word to the response.
02 / My contribution
Where I got
my hands dirty.
Orchestrating the conversation
Architected the FastAPI backend that brings transcription, streaming language-model inference, and voice generation into one production flow.
Making latency a product concern
Tuned async workflows under real-time video constraints, treating end-to-end responsiveness as part of the user experience.
Owning the full system
Joined as SimCare's first hire and worked across backend reliability, infrastructure, and Next.js product surfaces, helping shape the platform's technical direction.
03 / The engineering idea
The model is one part. The experience is the whole system.
A faster model alone doesn't create a natural conversation. The orchestration between services, the way responses stream, and the reliability of the backend all shape what the person on the other side feels.
What stays with me
Good AI engineering happens between the models, too. The connections are what turn individual capabilities into a product someone can use.
Artisan
AI automation · Backend systems