The Problem: Unstructured Data Overwhelming Clinical Staff

Approximately 80% of medical data exists in an unstructured format, like a doctor’s free-text note, while only 20% is stored in uniform, structured fields like lab values or demographics . This imbalance means highly trained oncology staff spend hours manually reading records to support essential workflows: clinical trial matching, patient visit preparation, and cancer registry reporting .

This manual overhead has real consequences. It creates a bottleneck in clinical trial enrollment—a process the National Cancer Institute says is vital for managing side effects and testing new treatments . It also leads to "pajama-time" hours where clinicians catch up on administrative work after shifts end . Triomics targets this exact pain point by automating the ingestion and reasoning across hundreds of pages per patient.

How OncoLLM Works: An Oncology-Specific AI Architecture

Triomics’ platform is powered by OncoLLM, an enterprise-grade AI framework purpose-built for oncology . Rather than a single monolithic model, OncoLLM is described as a constellation of 8 models ranging from 3 billion to 72 billion parameters that work in an agentic manner . This design allows the system to interpret information at the patient level, reasoning across a longitudinal medical record rather than analyzing one document at a time.

The technical approach is a deliberate departure from previous methods like named entity recognition or relation extraction . The company also leverages Microsoft’s Azure AI and OpenAI services, including finetuning the small language model Phi-3.5 to parse critical clinical information from unstructured data at scale . According to Microsoft, this integration allows the platform to review complete patient records against hundreds of ongoing clinical trials in under a minute .

Two core software products sit on top of OncoLLM:

• Prism: Automatically screens every upcoming patient appointment against an entire trial portfolio and delivers daily or weekly matches directly into the provider’s workflow, including an itemized, criterion-by-criterion rationale with source citations .
• Harmony: Curates EHR data to automatically assemble expert-level snapshots with current disease status, treatment timeline, biomarkers, and trial options . It also supports cohort analysis and precision oncology reporting .

In early validation with the Medical College of Wisconsin, OncoLLM reportedly found 90% of eligible patients for clinical trials in minutes—a task that would have taken qualified nurses days or weeks . The same source notes OncoLLM extracted structured data from unstructured notes at accuracy similar to or higher than models like GPT-4 or Claude, while being roughly 40 times cheaper to run .

Funding and Traction

The $22 million Series B follows a $15 million Series A raised in 2024 . The capital will be used to accelerate adoption across health systems and deepen integrations with electronic health records . The company does not publicly disclose detailed growth metrics like ARR or enterprise customer count, though the funding announcement positions it as trusted by leading cancer centers .

What’s publicly verifiable is the customer list. Triomics has secured deployment agreements with prestigious institutions:

• Memorial Sloan Kettering Cancer Center (MSK): Through iHub and CRIC programs, MSK is deploying the clinical trial matching and screening system across its active trials .
• Mount Sinai Tisch Cancer Center: Launched the PRISM platform powered by OncoLLM, becoming the first NCI-designated Comprehensive Cancer Center in NYC to deploy an oncology-specific AI tool for trial matching .
• Yale Cancer Center and Smilow Cancer Hospital: Expanded use of the platform across the entire portfolio of active oncology clinical trials, with results published in Nature Digital Medicine .

Competitive Positioning: Vertical Specialization vs. Generalist Tools

Triomics enters a market crowded with ambient AI scribes and clinical documentation tools such as Microsoft’s Nuance DAX Copilot and Abridge. However, its differentiation is vertical specificity.

General-purpose AI scribes are designed for broad clinical documentation across specialties—summarizing conversations between a doctor and patient during an appointment. In contrast, Triomics focuses exclusively on oncology workflows involving high-volume, long-form unstructured data that spans years of patient history . Its AI reads across the entire patient record, producing a structured, cited patient view before the visit, during screening, and after the appointment .

The company’s leadership also established the Collaboration for Oncology focused LLM Training (COLT), a consortium of over 20 NCI-designated cancer centers along with Ci4CC to create performance benchmarks and safety standards for generative AI in oncology . This positions Triomics as a stakeholder in setting the guardrails, not just building a product.

What’s Next for Oncology-Specific AI?

The $22 million funding raise signals a broader thesis taking hold in health tech: oncology’s data complexity demands specialized infrastructure, not just a repurposed general-purpose chatbot . As cancer centers face growing pressure to match more patients with precision therapies and clinical trials, the bottleneck is becoming a competitive necessity to solve.

Whether Triomics can sustain its first-mover advantage against well-funded generalists and enterprise EHR vendors remains an open question. But with deployments at MSK, Yale, and Mount Sinai, and a total $36 million raised, it has moved beyond proof-of-concept into scaled clinical operations. The next year will test whether vertical AI can deliver on its promise in one of medicine’s most data-intensive fields.

The Problem: Unstructured Data Overwhelming Clinical Staff

Approximately 80% of medical data exists in an unstructured format, like a doctor’s free-text note, while only 20% is stored in uniform, structured fields like lab values or demographics . This imbalance means highly trained oncology staff spend hours manually reading records to support essential workflows: clinical trial matching, patient visit preparation, and cancer registry reporting .

This manual overhead has real consequences. It creates a bottleneck in clinical trial enrollment—a process the National Cancer Institute says is vital for managing side effects and testing new treatments . It also leads to "pajama-time" hours where clinicians catch up on administrative work after shifts end . Triomics targets this exact pain point by automating the ingestion and reasoning across hundreds of pages per patient.

How OncoLLM Works: An Oncology-Specific AI Architecture

Triomics’ platform is powered by OncoLLM, an enterprise-grade AI framework purpose-built for oncology . Rather than a single monolithic model, OncoLLM is described as a constellation of 8 models ranging from 3 billion to 72 billion parameters that work in an agentic manner . This design allows the system to interpret information at the patient level, reasoning across a longitudinal medical record rather than analyzing one document at a time.

The technical approach is a deliberate departure from previous methods like named entity recognition or relation extraction . The company also leverages Microsoft’s Azure AI and OpenAI services, including finetuning the small language model Phi-3.5 to parse critical clinical information from unstructured data at scale . According to Microsoft, this integration allows the platform to review complete patient records against hundreds of ongoing clinical trials in under a minute .

Two core software products sit on top of OncoLLM:

• Prism: Automatically screens every upcoming patient appointment against an entire trial portfolio and delivers daily or weekly matches directly into the provider’s workflow, including an itemized, criterion-by-criterion rationale with source citations .
• Harmony: Curates EHR data to automatically assemble expert-level snapshots with current disease status, treatment timeline, biomarkers, and trial options . It also supports cohort analysis and precision oncology reporting .

In early validation with the Medical College of Wisconsin, OncoLLM reportedly found 90% of eligible patients for clinical trials in minutes—a task that would have taken qualified nurses days or weeks . The same source notes OncoLLM extracted structured data from unstructured notes at accuracy similar to or higher than models like GPT-4 or Claude, while being roughly 40 times cheaper to run .

Funding and Traction

The $22 million Series B follows a $15 million Series A raised in 2024 . The capital will be used to accelerate adoption across health systems and deepen integrations with electronic health records . The company does not publicly disclose detailed growth metrics like ARR or enterprise customer count, though the funding announcement positions it as trusted by leading cancer centers .

What’s publicly verifiable is the customer list. Triomics has secured deployment agreements with prestigious institutions:

• Memorial Sloan Kettering Cancer Center (MSK): Through iHub and CRIC programs, MSK is deploying the clinical trial matching and screening system across its active trials .
• Mount Sinai Tisch Cancer Center: Launched the PRISM platform powered by OncoLLM, becoming the first NCI-designated Comprehensive Cancer Center in NYC to deploy an oncology-specific AI tool for trial matching .
• Yale Cancer Center and Smilow Cancer Hospital: Expanded use of the platform across the entire portfolio of active oncology clinical trials, with results published in Nature Digital Medicine .

Competitive Positioning: Vertical Specialization vs. Generalist Tools

Triomics enters a market crowded with ambient AI scribes and clinical documentation tools such as Microsoft’s Nuance DAX Copilot and Abridge. However, its differentiation is vertical specificity.

General-purpose AI scribes are designed for broad clinical documentation across specialties—summarizing conversations between a doctor and patient during an appointment. In contrast, Triomics focuses exclusively on oncology workflows involving high-volume, long-form unstructured data that spans years of patient history . Its AI reads across the entire patient record, producing a structured, cited patient view before the visit, during screening, and after the appointment .

The company’s leadership also established the Collaboration for Oncology focused LLM Training (COLT), a consortium of over 20 NCI-designated cancer centers along with Ci4CC to create performance benchmarks and safety standards for generative AI in oncology . This positions Triomics as a stakeholder in setting the guardrails, not just building a product.

What’s Next for Oncology-Specific AI?

The $22 million funding raise signals a broader thesis taking hold in health tech: oncology’s data complexity demands specialized infrastructure, not just a repurposed general-purpose chatbot . As cancer centers face growing pressure to match more patients with precision therapies and clinical trials, the bottleneck is becoming a competitive necessity to solve.

Whether Triomics can sustain its first-mover advantage against well-funded generalists and enterprise EHR vendors remains an open question. But with deployments at MSK, Yale, and Mount Sinai, and a total $36 million raised, it has moved beyond proof-of-concept into scaled clinical operations. The next year will test whether vertical AI can deliver on its promise in one of medicine’s most data-intensive fields.