Palo Alto, California, United States · Hybrid
$170k–$230k/yr
Senior ML Engineer, Clinical Automation Hybrid in Palo Alto (Mon / Wed / Thurs) · $170K–$230K base + meaningful early-stage equity + full benefits About Sodalis Specialty pharmacy is where the most complex and expensive …
Skills: Machine learning, NLP, LLMs, Information extraction, Document understanding
Sign up with Clera and we'll reach out the moment a role actually fits you — no more spraying applications into the void.
You will own the development of AI agents and ML pipelines to automate clinical prescription reviews and data extraction. This involves building evaluation infrastructure, managing data flywheels, and ensuring models are explainable and trustworthy for pharmacists.
The role requires 5+ years of experience building production-grade ML systems, specifically in NLP, LLMs, and document understanding. Candidates must be pragmatic, comfortable with high-stakes data, and capable of owning end-to-end model lifecycles in a startup environment.
Specialty pharmacy is where the most complex and expensive drugs reach the most in-need patients, and it still runs on faxes, phone tag, and manual data entry. Patients wait weeks for therapies that can't wait.
Sodalis is building AI agents that run those operations end to end: reading and validating prescriptions, handling phone calls, and moving prescriptions through eligibility and fulfillment. The ambition goes beyond that. We want to be the rails every specialty prescription runs on, and the first system to perform clinical review at the standard of a pharmacist.
We're already in production, processing tens of thousands of prescriptions a week. Backed by Gradient Ventures (Google's AI fund), and founded by veterans of Apple's Applied ML, Included Health, and Avella Specialty Pharmacy (scaled to $1.5B+ before its acquisition by OptumRx).
Before a specialty prescription is dispensed, a pharmacist has to review it: the right drug and dose for this patient, interactions, contraindications, whether the therapy makes sense given everything else going on. That's the problem you'd own: AI that performs that same clinical review and presents it for verification. Not to replace clinical judgment, but to do the legwork so pharmacists can operate at the top of their license.
It's unsolved and genuinely hard. The system has to know what it knows and defer cleanly when it doesn't, earn a pharmacist's trust with reasoning they can verify rather than a black-box score, and be measured against clinical ground truth that no existing benchmark captures. Clinical experts label the data that establishes it, and you build the evaluation on top.
You'd own this and the intelligence layer beneath it: the models, pipelines, and eval infrastructure that turn messy clinical inputs into structured, trustworthy actions, plus the data flywheel that improves them with every prescription. It's the moat, and you'd be the ML engineer who owns it end to end, working directly with the founders.
Own extraction and structuring of messy clinical inputs (faxed prescriptions, prescriber directions, benefit data), turning free-form documents into accurate, structured data the pipeline can act on.
Build the clinical review that performs the checks a pharmacist would (interactions, contraindications, dose and therapy appropriateness), as explainable reasoning they can verify and trust.
Build the confidence scoring and routing that decides what's safe to automate and what needs a human in the loop.
Stand up the eval infrastructure, curated datasets, and production feedback loops that drive continuous accuracy gains: the data flywheel.
Choose the right tool for each problem, from prompting and fine-tuning LLMs to classical ML, biased toward what ships and holds up in production.
Partner closely with pharmacy operations and engineering to understand the real work, and turn what you learn into better models.
5+ years building ML systems that run in production: you've owned models end to end, not just trained them in a notebook.
Deep experience across information extraction, NLP, document understanding, and LLMs.
You own the machinery around the model: eval harnesses, data pipelines, and the infrastructure it runs on, and you improve it against real metrics.
Comfortable with messy, high-stakes, real-world data where correctness matters.
Pragmatic about models: you reach for the simplest approach that works, not the fanciest.
You've shipped 0→1 in startup environments and are comfortable owning ambiguity.
Curiosity about the clinical domain and eagerness to learn it deeply, working closely with pharmacists. No healthcare background required.
We're a handful of people going all in on a problem we think is worth it, holding a high bar and shipping work that reaches real patients. It's intense, and it isn't for everyone. We want people who want to build something that matters, alongside others who care as much as they do. If that's you, we should talk.
AI powered patient engagement and data entry for pharmacy
Sodalis uses pharmacy-specific generative AI to handle phone calls and text messages through natural, human-like conversations. With each interaction, it pushes outcomes directly into your pharmacy’s workflow through robust integrations. By increasing your labor capacity on demand, Sodalis enables faster pharmacy growth and frees up your staff to focus on high-value tasks.
Offices: Menlo Park, CA 94025, US
AI powered patient engagement and data entry for pharmacy
Sodalis uses pharmacy-specific generative AI to handle phone calls and text messages through natural, human-like conversations. With each interaction, it pushes outcomes directly into your pharmacy’s workflow through robust integrations. By increasing your labor capacity on demand, Sodalis enables faster pharmacy growth and frees up your staff to focus on high-value tasks.
Offices: Menlo Park, CA 94025, US
