Assort Health was founded in May 2023 by Jeffery Liu, a Duke computer science graduate who ran product engineering at Athelas after a stint at Facebook, and Jon Wang, a Stanford bioinformatics graduate who left UCSF medical school to build companies. The product is an AI phone agent that picks up when a patient calls a specialty practice, figures out which sub-specialist and visit type they need, and writes the booking into Epic, athenahealth, ModMed, NextGen or one of a dozen other practice systems. Funding came fast: a $3.5 million seed in 2024, a $22 million Series A led by First Round and Chemistry in April 2025 (the company’s own post counts $26 million raised to date, seed included), a $76 million Series B led by Lightspeed on September 30, 2025, and the $120 million Series C from Menlo Ventures on June 25, 2026, for $222 million total. TechCrunch put ARR at a little over $3 million in August 2025; job postings now claim revenue up 20x in 15 months and headcount growing from 15 in early 2025 to 250 by the end of 2026. Liu owns engineering, product and implementation under the co-CEO split, and LinkedIn lists Daniel Rothman as head of engineering. Do not confuse Assort with Infinitus, whose agents call payers on behalf of providers; Assort’s agents talk to patients.
Process
There is no published loop, but Glassdoor holds enough reports to sketch it. A Software Engineer candidate who interviewed in July 2025 describes a recruiter round, a take-home assignment with about a week to complete it, a live technical coding round over Google Meet, and an in-person final round in the San Francisco office mixing behavioral questions with technical questions about Python FastAPI and a mock EHR. The take-home was to build a voice agent; reference checks came after the onsite; difficulty was rated 3.5 out of 5. A separate Agent Software Engineer report counts about nine calls over six weeks: recruiter, a technical discussion, a system design interview, a coding interview, an onsite of four 30-minute conversations, and an offer call, with each stage explained in advance. Levels.fyi aggregates 24 reports at medium difficulty with only a third rating the experience positive, so expect variance between interviewers.
| Stage | Who | What it tests |
|---|---|---|
| Recruiter screen (30 min) | Talent team | Motivation for healthcare, seniority match, hybrid SF availability, comp expectations |
| Take-home (about one week) | Reviewed by an engineer | Build a working voice agent: speech in, LLM reasoning, tool calls, speech out, plus how you handle failure |
| Live coding (Google Meet) | Engineer | Practical Python, async patterns, API design; candidates report FastAPI and a mock EHR service |
| System design (Agent SWE track) | Senior engineer or engineering lead | Real-time call pipeline, EHR integration, latency and reliability under LLM provider failures |
| Onsite in SF (four 30-min sessions) | Engineers, product, a founder | Behavioral depth, technical follow-ups on your take-home, product judgment, culture |
| References and offer call | Recruiter, hiring manager | Two to three references, then band placement and options grant |
Two warnings. Founder interviews are part of the loop for some roles, and a Blind post from summer 2026 describes a case-study interview with a founder who rejected the candidate on the spot with blunt feedback; an Assort representative apologized in the thread. Treat the founder round as a real evaluation of how you think about the customer. And the loop runs long by early-stage standards: nine touchpoints over six weeks is more than Hippocratic AI or most peers in our AI startup difficulty index, so ask the recruiter how many stages your track has and whether any can be combined.
What they actually ask
- Build a voice agent that can take a call, understand a patient asking to book a follow-up with a knee surgeon, and confirm a slot. Walk us through what you built for the take-home, why you chose that speech-to-text and text-to-speech stack, and what happens when the LLM returns garbage mid-sentence.
- A patient is on the line and the LLM provider starts returning 429s. You cannot drop the call. What does your retry, fallback and graceful-degradation path look like, and how do you keep the caller from hearing silence?
- Design the scheduling layer for a practice with 40 providers, each with their own rules about which visit types they accept, which insurances they take, and which sub-specialty they see. How do you represent those rules, keep them in sync with athenahealth, and stop the agent from booking something the front desk would have refused?
- Here is a mock EHR with a FastAPI interface. Write the endpoints an agent would need to look up a patient by phone number and date of birth, list open slots, and create an appointment, and tell us how you would cache slot availability without serving stale data.
- How would you evaluate whether a new prompt or model version made the agent better or worse across 62,000 care protocols? What is the regression suite, what is the metric, and how do you catch a change that helps orthopedics but breaks dermatology?
- Call transcripts contain names, dates of birth and diagnoses. Describe how you would log enough to debug a bad call while keeping PHI out of your observability stack and out of the LLM vendor’s training data.
- Tell us about a time you shipped something a customer did not want. What did you learn about the gap between what they asked for and what they needed?
Levels and comp (2026)
The Ashby job board publishes base bands for every San Francisco engineering role, all hybrid and all marked as offering equity. As of September 2026 the Senior Software Engineer listing (4 to 8+ years) shows $190K to $240K, Staff Software Engineer (7+ years) $240K to $295K, Forward Deployed Engineer $155K to $210K, Senior Forward Deployed Engineer $190K to $240K, and Engineering Manager and Head of Forward Deployed Engineering $210K to $300K. Earlier Glassdoor-mirrored listings for the Agent Software Engineer ladder ran $155K to $185K at entry and $190K to $220K at senior, so the bands moved up after the Series C. Glassdoor’s salary page has 16 self-reported entries and levels.fyi has no verified offers, so the posted bands are the best signal available. Options are standard, with no public data on grant size or strike price. At a $1.2 billion preferred valuation three years after founding, common-stock options sit under $222 million of liquidation preference, and a flat exit would compress them hard. Ask for the fully diluted share count and the last 409A, negotiate base against the top of the band, and treat equity as upside rather than salary.
Prep priorities
Learn the domain first, from the front desk’s point of view: visit types, provider templates, referral requirements, insurance eligibility checks, and why an orthopedic group refuses to let a hand surgeon see a hip patient. Then learn the plumbing an agent needs: the athenahealth and Epic FHIR APIs for patient lookup and booking, what a Twilio or SIP media stream looks like, and the difference between streaming speech-to-text latency and end-to-end turn latency. Sharpen async Python (the live round has surfaced FastAPI), build one small voice agent end to end before you apply so the take-home is a refinement rather than a first attempt, and rehearse a system design answer for a real-time call pipeline that survives an LLM outage. Infrastructure postings name GCP, Kubernetes, Grafana and Prometheus, so be ready to discuss observability for a system whose failure mode is a patient hearing dead air.
Ask the recruiter which track you are on, since the Software Engineer and Agent Software Engineer loops differ in length and in whether a take-home is used. Ask who runs the onsite and whether a founder round is included, and how many office days the hybrid label means. With the company adding roughly 200 people in a year, leveling is being invented in real time, and the gap between senior and staff is worth more than the base salary difference.
Frequently Asked Questions
Does Assort Health require a take-home assignment?
For the Software Engineer track, at least one candidate reports a take-home with about a week to complete it, and the assignment was to build a voice agent. The Agent Software Engineer report describes live coding and system design rounds instead. Ask the recruiter which applies to you before you commit the week.
Is Assort Health remote or in-office?
Every engineering listing on the Ashby board is tagged San Francisco and hybrid, and the final round is held in person at the SF office. Only a couple of sales and marketing roles are remote. If you need full remote, the engineering postings are not written for you.
How long does the Assort Health interview process take?
One Agent Software Engineer candidate counted nine calls over six weeks from recruiter to offer. The Software Engineer track adds a week-long take-home plus reference checks, so plan on four to six weeks. A competing deadline usually speeds things up.
How is Assort Health different from Infinitus or Abridge?
Assort’s agents answer inbound calls from patients and place outbound calls to them, handling scheduling, intake, refills and referrals for specialty practices. Infinitus points its voice agents at insurers and pharmacies on behalf of providers, and Abridge records the clinical visit and writes the note. TechCrunch listed EliseAI, Hello Patient, Abridge and Ambience as the companies Assort competes with for capital; the closest product overlap is with Hello Patient.
Similar company guides
Prepping for Assort Health? Put it to work:
