Our verdict
Xyonix is worth hiring for a startup or mid-sized company, often in healthcare, that wants senior machine learning engineers to check whether its data can support a model and then build it, at published rates of $200 to $300 an hour. It is a very small firm with little recent outside review evidence, so it suits one well-scoped model or product feature better than a large program.
The right buyer
- Health and health tech companies that need a model built from messy text, video or audio. For DexCare it built a 150-label ontology of patient reasons for visit and shipped the parser in a Docker container (DexCare testimonial).
- Companies unsure whether their data can support a model. Its Pathfinder phase reviews the data and goals first; for Delta Dental of Washington it lasted two weeks (Clutch).
- Startups testing a product idea that depends on machine learning. The CTO of Madrona Venture Labs told Clutch the firm is “pragmatic as well as affordable” and had a working proof of concept up quickly.
- Buyers who want to know the price range early: Clutch lists $200 to $300 an hour and a $5,000 minimum project.
Reasons to look elsewhere
- It is very small. Clutch puts its headcount at 2 to 9 employees, so it cannot staff several projects or a large team at once.
- Outside evidence is thin and dated. It has two verified Clutch reviews, both from 2021; most other client accounts are testimonials on its own site.
- Not every project lands. Delta Dental of Washington said its first phase did not succeed before a second phase reached the 70% model accuracy it wanted (Clutch).
- It can be too patient with a slow client. Xometry’s technology chief said he wished the firm had pushed harder for a decision when his own team stalled (Xometry testimonial).
Xyonix, At A High Level
Xyonix is an AI development, implementation and consulting firm in Seattle, Washington, that builds custom machine learning models and AI features for startups and mid-sized companies.
- Founded in 2016 (Clutch) by Deep Dhillon and Carsten Tusk, who worked together at the language search startup Evri (Team) and share several US patents (Justia Patents).
- A small senior team of 2 to 9 employees, according to Clutch; its team page shows the two founders plus a partner and principal data scientists.
- Work starts with a discovery phase it calls Pathfinder, then moves to model building, hosting and handoff. Hosted models can run on its own open-source Mayetrix platform, and clients have received tested Docker containers and code at handoff (case study).
- Clutch shows its AI work split mainly between machine learning and natural language processing, and puts 60% of its clients in the $10 million to $1 billion revenue range.
- Hosts the Your AI Injection podcast, in which Deep Dhillon interviews people building AI products (Xyonix).
Its documented client work includes:
- A reason-for-visit parser for DexCare, a health care access software company: Xyonix reviewed tens of thousands of patient entries, built a 150-label ontology, trained labelers and shipped the model in a Docker container for DexCare’s own Kubernetes cluster (DexCare testimonial).
- Models for Xometry, a manufacturing marketplace, to help customers move from a quote to an order and to explain why some manufacturing partners stay active and others leave. Its CTO said he talked to 15 to 20 firms before choosing Xyonix (Xometry testimonial).
- Member retention models and an interactive plan finder for Delta Dental of Washington, a dental insurer, under an ongoing retainer that started in November 2019 (Clutch review; plan finder testimonial).
- Video and text models for a robotic surgery training startup that point out specific surgical moves in hours-long videos; the system was handed to the client’s staff after a Fortune 500 company acquired the startup (case study).
- A prototype that forecasts hourly patient arrivals at hospital units up to five days ahead for a patient management software company, with SHAP explanations for each forecast (case study).
Scores
| Criterion | Weight | Score | Basis |
|---|---|---|---|
| Production track record | 20% | 6.5 | DexCare parser running in the client’s own cluster; surgery video models in use after the client’s acquisition; Delta Dental models under retainer; several other projects described as prototypes. |
| Technical depth | 15% | 7.5 | Founders hold US patents in language search, entity recognition and image search; work spans text, video, audio and time series models. |
| Industry depth | 10% | 6.5 | Repeat work in health care and health tech, plus insurance, education, media and manufacturing; no single industry dominates. |
| Delivery model and handoff | 15% | 7.0 | Pathfinder discovery phase, regular review meetings and handoff with Docker containers, code and staff training. |
| Commercial clarity | 10% | 7.0 | Clutch lists $200 to $300 an hour and a $5,000 minimum; the most common reviewed project ran $50,000 to $199,999. |
| Platform independence | 10% | 7.0 | Builds on AWS and open-source tools and works with several model providers; optional hosting on its own open-source Mayetrix platform. |
| Stability and scale | 10% | 4.5 | Operating since 2016 with 2 to 9 employees; too small for large or parallel programs. |
| Independent evidence | 10% | 5.5 | Two verified Clutch reviews from 2021 averaging 4.8; client-attributed testimonials on its own site; little press coverage. |
| Overall | 100% | 6.5 | Weighted average |
Where Xyonix fits
| Need | Fit | Why |
|---|---|---|
| Custom models from clinical text, video or audio | Strong | DexCare reason-for-visit parser; robotic surgery video models; patient arrivals forecaster. |
| Testing whether data can support a model before a full build | Strong | Pathfinder phase used at Delta Dental of Washington and the surgery startup. |
| Early machine learning product work for startups | Good | Madrona Venture Labs proof of concept on AWS; Clutch reviewer called it pragmatic and affordable. |
| Large programs needing many engineers at once | Weak | 2 to 9 employees. |
What clients and leaders say
“Xyonix’s ability to take a problem that has very little structure around it, build an ontology, and provide an end-to-end solution was pretty impressive.”
“Deep and Bill never tried to handwave or paint rosy pictures - if they thought something was going to be hard, they were vocal about it.”
“When we were stumbling a bit on our side it would have been nice if Deep would have put a little more pressure on us and say are you in or out instead of being too patient.”
Who leads Xyonix
Sources (12)
- Xyonix: homepage
- Xyonix: Team
- Xyonix: Customer projects
- Xyonix: AI backed surgery case study
- Xyonix: Patient arrivals forecaster case study
- Xyonix: AI med spa assistant case study (Mayetrix and handoff)
- Xyonix: DexCare testimonial (Eric Kolve)
- Xyonix: Xometry testimonial (Matt Leibel)
- Xyonix: Delta Dental of Washington plan finder testimonial
- Clutch: Xyonix profile and verified reviews
- Justia Patents: Carsten Tusk patents (with Navdeep S. Dhillon)
- Great Things with Great Tech: Deep Dhillon interview (March 2023)