Skip to the work
Shrivathsan S.
← Shrivathsan Sakthisundaram

Stop 4 · the therapy office

Clinical notes that cite the record, adopted by 20+ therapists.

AphasiaGPT

Founding Engineer · Nov 2025 to Apr 2026 · Durham

AphasiaGPT, drawn as the therapy office in the townAphasiaGPT, drawn as the therapy office in the town

Clinical documentation software for speech-language pathologists.

The problem
Speech-language pathologists wrote therapy notes from memory after sessions.
What I did
Built cited retrieval on PostgreSQL full-text search around their SOAP note workflow. Then traced a primary-key mismatch that emptied every note lookup and fixed it across 14 call sites in 8 tables.
The result
Adopted by 20+ therapists with 50+ patients; 47 merged full-stack PRs.
therapists adopted it
20+therapists adopted it
patients
50+patients
merged full-stack PRs
47merged full-stack PRs

The problem

Speech-language pathologists wrote therapy notes from memory after sessions. I interviewed 5+ SLPs and designed around the SOAP note workflow they already used.

What I built

Per-patient note retrieval on PostgreSQL full-text search: a weighted tsvector, a GIN index, and inline citations back to the source notes. I later extended it with embeddings.

47 merged PRs across the web portal and 13 back-end edge functions. I shipped note-taking and SOAP generation on a deadline, before a national aphasia technology talk where the beta was demoed.

What broke and how I found it

Every note lookup came back empty, with no error. I traced it to a silent primary-key mismatch and corrected 14 call sites across 8 tables. Then I repaired 6 schema columns that had drifted and were aborting deploys.

Result

Adopted by 20+ therapists with 50+ patients.