Senior Software Engineer, Dermatology Inference

Specialty care is hard to access and even harder to run. Patients often wait months, and clinics rely on fragmented tools and manual workflows.

Tono builds software and operations together to make specialty care faster to access and easier to deliver, starting with dermatology.

We’re building a fully autonomous specialty medicine clinic: continuous, real-time care for patients anywhere.

What makes us different is how we build. We run two clinics of our own, a national telehealth practice and an in-person clinic in DUMBO, Brooklyn. Our engineers work side by side with our doctors and care coordinators to build (mostly) every piece of the software from scratch. The long-term goal is for specialty clinics across the country to run on our autonomous EHR, freeing them from the administrative burden that slows care down.

Today we see patients in 33 states, covering more than 80% of the US population.

You will work closely with the co-founders and a team of engineers and physicians (backgrounds include Amazon, Apple, and leading academic medical centers). We are backed by Village Global (backed by Mark Zuckerberg, Jeff Bezos, Michael Dell, chaired by Reid Hoffman), HOF Capital, and Oncology Ventures.

Build the specialized models that help our dermatologists see more, catch change earlier and decide faster. You'll own the full path from research idea to production model: developing, training, evaluating and deploying models for dermatology, including image-based diagnosis support and longitudinal lesion detection and tracking across a patient's visits over time. You'll work side by side with our physicians to define what a useful, trustworthy result looks like, and with the engineers who build our EHR and patient portal to get it in front of doctors.

On any given week you might be:

• Training and fine-tuning vision and multimodal models (including LoRA and other adaptation methods) on dermatology images, and deciding when to build custom versus adapt an open source model

• Building longitudinal pipelines that align and compare lesions across photos taken weeks or months apart, handling differences in lighting, angle, camera and skin tone

• Building labeled datasets and annotation workflows with clinicians, and using them to evaluate models honestly

• Standing up inference services: model serving, batching, latency and cost

• Adding confidence scoring, uncertainty estimation and human-in-the-loop review so doctors are never asked to trust a black box

• Monitoring models in production for drift, image quality problems and failure cases, with performance broken out across skin tones and patient groups

• Integrating results into our TypeScript product (tRPC, Prisma, Postgres) so output fits naturally into the doctor's workflow

Patients and doctors deserve better than a demo. We ship in a regulated environment with careful reviews, feature flags and safe rollouts.

• We ship quickly, with discipline: small PRs, clear ownership, and tight feedback loops.

• Small team, broad scope. Engineers regularly work across frontend, backend, data, and infrastructure.

• We ship with confidence in a regulated environment: feature flags, careful reviews, and safe migrations (including backfills). We use AI all the time but we do not vibe code our way into debt. Patients and doctors deserve better.

• We invest in maintainability. Dependency upgrades, performance work, and removing dead code are part of the roadmap.

• Observability is built-in. Logs, metrics, alerting, and error monitoring are part of the job.

• We value builders. Curiosity, good taste, and a willingness to pick up whatever the system needs goes a long way.

Frontend: Next.js, TanStack Start, React

Backend: Fastify, tRPC, Prisma, PostgreSQL

Durable Workflows: Inngest

AI: Mastra, OpenAI, Anthropic, Vertex

Infra: Cloudflare, Turborepo, Doppler, AWS, GCP

Observability: Posthog

Data: Motherduck/DuckDB, dbt, dlt, Prefect, BigQuery

• 5+ years of software engineering experience, including training and shipping ML or computer vision models into production

• Strong Python and deep learning experience (PyTorch or similar), with hands-on work in computer vision such as classification, detection, segmentation or image registration

• Experience with fine-tuning and adaptation methods (LoRA, transfer learning) on vision or multimodal models

• A rigorous approach to evaluation: clinically meaningful metrics, calibration, held-out and external validation sets, and an awareness of bias and failure across skin tones and imaging conditions

• Experience building and serving models in production: GPU workloads, inference performance, cost and monitoring

• Experience building datasets and labeling workflows with domain experts

• Comfort with relational databases, data pipelines (we use tools like dbt, DuckDB and BigQuery), and shipping iteratively with feature flags and instrumentation

• Comfort working in a TypeScript product codebase, or eagerness to learn it

• Care about patient safety, privacy and reliability, and take pride in your craft

• Nice to have: dermatology or medical imaging experience, longitudinal or change-detection models, published research or open source work in vision, regulated environments (HIPAA, FDA SaMD), startups, or experience working directly with clinicians

Salary: 180k - 325k commensurate with experience

Equity: Meaningful equity in a high-growth, venture-backed company

Medical: Medical, dental, and vision coverage

Growth: Direct access to founders and an agile team where your impact is highly visible. We do weekly demos with leadership

PTO: Unlimited PTO to recharge when you need it

New York City preferred. We are hybrid and currently spend a few days per week in our DUMBO, Brooklyn office. We will consider fully remote for the right candidate.