Stethy
Enterprise email-triage SaaS for top-tier global pharma. Multi-account AWS CDK, Phi-3.5 fine-tune via Unsloth, Langfuse observability.
The Problem
Pharma operations teams drown in high-volume internal email tied to quality events, adverse-event reports, and regulatory workflows. Off-the-shelf classifiers aren't domain-tuned; generic LLMs hallucinate on pharma jargon; and regulated QMS tooling doesn't speak email. Stethy's invite-only pharma clients needed a single end-to-end pipeline: Gmail ingest, domain-tuned classification, structured QMS-compatible records, and observable LLM behaviour, without leaking regulated information to public LLM providers.
The Solution
Apex36 delivered a single enterprise SaaS spanning eight coordinated workstreams: the core workflow platform (FastAPI + SQLAlchemy + React) with a Chrome extension ingesting Gmail, a domain-tuned Phi-3.5 email classifier fine-tuned 2× faster via Unsloth, a multi-account AWS CDK infrastructure (dev/stage/prod) with Lambda + CodePipeline, Langfuse observability hosted on Azure Container Apps via Bicep, a TrackWise-compatible QMS demo, and active LLM RAG research using Quivr, Ragflow, LiteLLM, and AWS AgentCore.
Features
FastAPI workflow SaaS
Python + FastAPI + SQLAlchemy + Alembic backend with pytest coverage; React + Vite + TypeScript frontend, the core B2B invite-only email-triage platform.
Multi-account AWS CDK infrastructure
AWS CDK in Python across dev / stage / prod accounts with AWS Lambda and CodePipeline, a six-repo infra fleet under one client.
Phi-3.5 fine-tune via Unsloth
Domain-tuned email classifier on pharma email corpora, trained 2× faster via Unsloth and benchmarked on a 9-category / 150-sample test set.
Langfuse LLM observability
Langfuse hosted on Azure Container Apps via Bicep: every LLM call across the stack is observable.
TrackWise-compatible QMS demo
TypeScript + Next.js demo replicating Sparta Systems' TrackWise Digital QMS workflows so quality teams can see the email-to-QMS hand-off end to end.
Gmail Chrome extension
Chrome extension feeding corporate Gmail inboxes into the Stethy triage workflow.
LLM RAG research workstream
Active pharma-internal knowledge-base experiment using Python, Quivr, Ragflow, LiteLLM gateway routing, and AWS AgentCore.
Results / Impact
served as invite-only enterprise clients.
via Unsloth, benchmarked on a 9-category / 150-sample test set.
dev / stage / prod running a six-repo infra fleet under one client.
on every LLM call, hosted on Azure Container Apps via Bicep.
FAQ
Ready to build something impactful?
Let's discuss your project and how we can help you ship faster and smarter.
Book a Free Strategy Call