Working at Triomics: Culture, Pace and Who Thrives

The Rhythm of Work

Triomics quadrupled enterprise customers, TechCrunch reported, and grew annual recurring revenue tenfold, according to TechCrunch, in the past year, yet its 76-person team split between New York and Bengaluru operates with a deliberate hiring mandate. This profile examines the day-to-day reality of working at Triomics: how work gets done, the operating principles that guide decisions, and what the hiring bar selects for.

The company operates as a distributed organization with roughly 76 employees as of 2026. U.S. staff combine remote and on-site time; the India team anchors the engineering center in Bengaluru. The split is deliberate: AshbyHQ postings show core platform and infrastructure roles in India, customer-facing deployment and go-to-market roles in New York. The careers page frames it explicitly: build the operating rhythms that let a distributed U.S. and India team function as one, including hours overlap, communication norms, and managerial structures that hold up across time zones.

Decision-making authority sits with the people doing the work. "Decisions get made by the people doing the work" isn't a slogan — it's the operating principle. Small teams own outcomes end to end. The feedback loop is short by design: clinical fellows, registrars, physicians, and informaticists from partner centers like Memorial Sloan Kettering and Yale Cancer Center sit in standups and user research sessions. MSK's Senior VP for Clinical Research, Paul Sabbatini, joined Triomics' Customer Advisory Board to guide scale-up across the network.

Communication is written by default. "Most of our internal communication is written. Most of our external work is documents, papers, and citations. People who think and write clearly thrive here." The same rigor applies internally: bias toward integrity over speed, calibrated confidence over velocity, willingness to say "no" when evidence doesn't support a claim. "We'd rather slow down to be right than ship a confident wrong."

The pace is "calm, durable," a long-cycle problem where urgency is reserved for moments that matter, not manufactured for theater. Compensation and leveling frameworks, performance management, and HR operations are being built to hold across two countries and multiple time zones.

What Guides Decisions

Triomics builds its operating philosophy around a single observation: oncology teams drown in unstructured data while patients miss trials that could help them. "We have seen medical records with thousands of pages of information," co-founder Sarim Khan told TechCrunch in May 2026. The company's response isn't to layer another tool on top of the chaos but to normalize data at the patient level, then reason across the whole record. That principle — patient-level reasoning over document-level summarization — shapes every product decision, from the ONCO LLM architecture to the insistence that each institution gets its own model instance with no cross-customer data sharing.

The company publishes its north star on its About page: "Every clinician should walk into the visit already informed. Every patient should have a fair shot at finding the right trial. Every quality reporting team should start with trusted, structured data." Internally, they measure against a harder metric: 100 percent of scheduled patients screened against every active trial, every day, with matches cited back to pathology, molecular, and note-level evidence. Khan told CHAI in October 2025 that even at large academic centers, only 20-30% of patients were being screened manually before Triomics deployed. The gap between upwards of 70% of patients saying yes to trials and only around 5-7% actually enrolling is, in their view, largely an identification problem, one that compounds as inclusion criteria grow more complex.

When a health system onboards, it receives a dedicated ONCO LLM instance. User interactions and refinements stay within that tenant. No patient data, no model weights trained on one institution's records, move to another. This design choice reflects a deliberate bet that academic medical centers will not adopt generative AI without ironclad data boundaries. The trade-off appears in the company's own CHAI interview: "It's been a roller coaster with larger academic research centers and health systems as they actively try to define their own AI guidelines and policies while deploying technology like ours. You can iterate on requirements, and three weeks later there's a whole new list."

Validation follows similar rigor. Triomics has published in Nature Digital Medicine with academic partners, reporting 95% accuracy in patient-to-trial matching via the PRISM system. The company cites the same figure in its ASCO 2025 abstract and Forbes coverage from May 2024. But the operating principle isn't the headline number; it's the insistence on citation. Every match, every pre-charting note, every registry abstraction links back to the source line in the chart. Clinicians do not trust black boxes; they verify. The product is built for that verification loop.

Collaboration over silos appears as a stated value in the company's CHAI membership rationale. "We joined CHAI because we wanted to provide back those learnings and be part of the team that helps set reliable roadmaps for deployment going forward," Khan said. The same interview frames the current moment as the first few innings of enterprise AI deployment at scale and argues that sharing deployment learnings across vendors, academic centers, and community oncology is the only way the ecosystem moves past pilot purgatory. That stance (publishing methods, joining standards bodies, treating competitors as potential co-authors on the infrastructure layer) distinguishes Triomics from the scribe vendors (Abridge, Nuance) it is often grouped with. Those tools summarize conversations; Triomics structures the longitudinal oncology record.

The hiring signal follows from these principles. The company recruits forward-deployed ML engineers who can sit inside a cancer center's workflow, site reliability engineers who treat uptime as a patient-safety issue, and account managers fluent in both NCI designation requirements and registry abstraction specs.

The friction lives in the pace of institutional change. Health systems rewrite AI policies monthly. Integration queues stretch. The company's own LinkedIn commentary acknowledges that solving the data layer — matching patients to trials — exposes the next bottleneck: the operations layer (CTMS, EDC, eTMF, IRT) that still runs on manual handoffs.

Inside the Interview Loop

Triomics runs a compressed but thorough loop: three rounds for software engineers, four for data scientists, each spanning roughly three to five weeks. The recruiter screen filters for baseline fit and completed applications; incomplete forms get auto-rejected before a human sees them. From there the process splits by discipline but converges on the same signal: technical depth in a regulated, data-messy, high-stakes clinical environment.

Technical competency is table stakes, but the bar sits higher than "can you code." Software-engineer candidates face deep dives into their primary language (Python or TypeScript) plus the internals of the frameworks they claim. Interviewers probe performance implications, library internals (PyTorch, HuggingFace), and how you structure data when integrity is non-negotiable. Data Structures & Algorithms scores 99 on Interview Query's frequency index; Machine Learning (75) and SQL (74) follow close behind. Probability (34) and Product Sense & Metrics (31) round out the top five. You will whiteboard. You will defend every design choice against alternatives you rejected. Silence when stuck is a negative signal; thinking out loud and taking hints is expected.

For data scientists, the loop adds causal inference, A/B test design, and production deployment experience. Must-haves: expert Python and SQL, statistics rigor, a track record of moving product strategy with data. Nice-to-haves tilt toward operations research (Gurobi, OR-Tools) and marketplace or logistics backgrounds; the research notes "complex environments such as logistics, maintenance, and marketplace operations" as the daily reality. Advanced degrees (MSc/PhD) appear on the nice-to-have list but do not gatekeep; proven impact does.

What separates an offer from a polite pass is the "why" and "so what." Interviewers explicitly look for candidates who connect technical decisions to business outcomes and articulate trade-offs: speed versus accuracy, complexity versus maintainability. Behavioral rounds (two for data scientists, one managerial round for engineers) test ambiguity handling, disagreement resolution, and learning agility in a field where the GenAI baseline shifts monthly. STAR-structured answers win; rambling narratives lose.

Domain curiosity carries unexpected weight. "Show genuine interest in healthcare AI, not just the technology" appears in the interview guides as a counter-move. The company's own careers page frames the mission bluntly: "If you want your work to matter to clinicians, researchers, and patients, read on."

Referrals correlate with a smoother, more structured experience per candidate reports. Cold applicants should budget two to four weeks of focused prep: SQL window functions, core stats, DS&A fundamentals, and a rehearsed project defense that covers data integrity guards and bias checks.

Seven salaried roles, median $150k, band $92k–$203k.

The Evidence Gap

Public employee-review data for Triomics exists but is limited. Glassdoor shows 12 reviews total; AmbitionBox shows 10+ reviews with a 4.0/5 rating. That volume reflects a company of 76 employees (per BuiltIn) operating in a niche — AI for oncology clinical-trial matching — that attracts a self-selecting talent pool. Candidates should treat the review volume as a data point, not a comprehensive signal.

The only grounded, first-party labor-market signal comes from Zero G Talent's job board, which shows the seven roles above posted recently. The spread suggests a company moving from product validation into scaled delivery.

No named current or former employees appear in the research record with on-the-record commentary about culture, management, or work-life balance. Founder interviews (Sarim Khan, co-founder and CEO; Hrituraj Singh, co-founder and CTO) discuss product vision and fundraising (Triomics raised a $15M Series A led by Lightspeed Venture Partners in 2024) but do not address internal operating norms or employee sentiment. Without attributed quotes or verified review excerpts, any characterization of "praise" or "criticism" would be fabrication.

What can be inferred from the hiring pattern is structural: a Bengaluru-based Director of Engineering role at a 10–20M INR band implies significant technical leadership authority seated in India, while the New York roles cluster around customer deployment and revenue functions. That geographic split often creates friction (time-zone handoffs, divergent engineering vs. commercial priorities, communication overhead) but the research contains no employee accounts confirming or denying it.

Candidates should ask directly in late-stage interviews: "How are decisions made when the engineering lead in Bengaluru and the forward-deployed team in New York disagree on priority?" and "What does a typical week look like for the SRE and the Forward Deployed ML Engineer right now?" The answers will be more reliable than any aggregated review score.

Working in frontier tech? Zero G Talent tracks the openings: see every open Triomics role, browse frontier tech jobs, the companies hiring, and the people building the field.