Detail
Patientdesk
The control plane is TypeScript and Next.js; the voice runtime is Python and FastAPI. Both run in two regions, US and Australia, against separate Supabase projects. The agents answer calls, book and reschedule appointments, check insurance and take payments, writing into OpenDental, CareStack, CorePractice, Dental4Windows and NexHealth. I built the call observability, the tenant isolation and encryption, and the deployment flow that moved clinic onboarding to self-serve.
TypeScript · Next.js · Python · FastAPI · Supabase · PostgreSQL · GCP · LiveKit · ElevenLabs · Stripe
ImpliedOptions
Options analytics on web, iOS and Android. Python services read the options tape, classify trades as blocks, sweeps or multi-leg strategies, add Greeks and open interest, and write history into partitioned Postgres so long lookbacks stay fast. On top of that: IV rank and percentile, gamma exposure, expected moves, max pain, Black–Scholes P&L and backtesting, served over streaming APIs with push alerts.
Python · FastAPI · PostgreSQL · Docker · Next.js · React
VoiceData
A marketplace where people are paid to record voice data. Contributors apply, record in the browser, on mobile or over a LiveKit call; submissions go through automated QA and human review; approved work is paid out weekly. It supplies the training and evaluation data for the model work at Patientdesk Labs.
Next.js · Supabase · LiveKit · PyTorch · Transformers · TRL · PEFT
News relevancy classifiers
A pair of BERT models that score a news headline for relevance on a four-point scale, so an aggregation pipeline can filter and rank without manual triage. FinBERT-ft-v3 fine-tunes TurkuNLP's Finnish BERT on ~225 hand-labelled Finnish headlines; BERT-ft-v2 does the same for English healthcare and AI news over ~254 headlines. Both are research artefacts, released on Hugging Face under MIT.
Python · PyTorch · Transformers · Hugging Face
Patientdesk.ai
Patientdesk Labs
VoiceData
FinBERT-ft-v3
Institute of Economic Affairs