Pharox DX unifies everything that exists about a case —diagnosis, biomarkers, pathology, imaging, labs and trials— into one map with explicit relationships between every element. From that map the physician decides, the patient understands, and Latin America's precision oncology ecosystem connects when the case calls for it.
In Latin America, an oncology patient's case does not exist anywhere as a coherent model.
It exists as documents scattered across institutions, formats and systems that do not talk to each other. The oncologist reconstructs it from memory at every visit; the patient carries it in physical folders. Neither has a complete, structured, up-to-date view of the case.
Molecular profiling may sit in a scanned PDF nobody ever structured; treatment history may exist only on paper. This is not a problem of will: it is one of infrastructure. Without a layer that unifies the case, no applied intelligence can reason over it.
With no access to the previous result, the study is simply ordered again — with the cost that implies and the toll it takes on the patient. Scattered information delays therapeutic decisions by weeks or months.
Clinical trials the patient might be eligible for that nobody assesses because nobody holds the full profile. Updated protocols that would change the treatment plan and that the oncologist had no time to review.
The anxiety of an oncology patient has many sources, and one of the most underestimated is not knowing what they have, what it means, what is missing and what comes next. You address it by building the map that organizes what already exists.
1 : 3,000
One oncologist for every 3,000 new cases a year in Argentina
Every minute of a visit spent reconstructing the case is a minute not spent deciding the treatment.
A persistent data model with explicit relationships between diagnosis, biomarkers, treatments, imaging, history and progression. It is not a document repository and not a dashboard: it is a graph that can be queried, reasoned over and updated — and it holds the memory of the whole case, not just its current state.
Reference case
María S., 54 years old
Invasive ductal carcinoma of the breast, cT2 cN1, clinical stage IIB. ER+ 90%, PR+ 70%, Ki-67 25%, grade 2. HER2 IHC 2+ with no confirmatory FISH.
IDC grade 2, no lymphovascular invasion, IHC complete
ER+, PR+ and Ki-67 confirmed — HER2 IHC 2+ without FISH, BRCA1/2 not ordered
Mammography BI-RADS 5 — staging CT pending
CBC, hepatic and renal function, CA 15-3, CEA
CDK4/6 + AI meets criteria (subject to menopausal status) — T-DXd and olaparib conditional
NCCN Breast v2.2026 guideline current for this profile
What matters is not the list of options: it is that the map knows which pending node turns each “conditional” into “meets criteria.” A FISH resolves T-DXd eligibility; a BRCA panel resolves olaparib; ECOG still needs to be recorded.
The map is persistent: it is not recomputed at every visit, it is updated with every new data point. When María's FISH arrives, the relationships change and the physician sees the updated case the next time they open it. The system shows the connections; the oncologist decides what to explore.
Each clinical area has an expert agent with its own knowledge domain, its own tools and the optimal model for its task. An orchestrator coordinates execution, manages dependencies between agents and synthesizes their findings into one coherent view.
Variants, FISH, HER2, ER, PR, Ki-67, BRCA and genomic panel. The highest-impact agent on the map: many treatment and trial nodes depend on the molecular profile.
Histological type, grade, lymphovascular invasion, margins and immunohistochemistry.
Mammography, ultrasound, CT, MRI and PET-CT. Structures findings according to BI-RADS and RECIST criteria.
CBC, hepatic and renal function, tumor markers (CA 15-3, CEA). Detects out-of-range values and trends over time.
Evaluates the full map against ClinicalTrials.gov and current guidelines, with continuous surveillance of their updates.
A rule such as “every patient with breast IDC and HER2 IHC 2+ has a pending confirmation node that conditions T-DXd eligibility” cannot be inferred by a language model: oncologists define and maintain it. The system executes those rules; it does not author them. Physicians trust a system that applies rules they validated.
The state of one node conditions how the other agents reason. If the imaging agent reports a complete response to neoadjuvant therapy, the stage changes in the map and the trials agent recomputes eligibility. If the blood agent reports neutropenia, that finding is incorporated as a temporary exclusion criterion. Every agent reads the map as updated by the others before reasoning.
The trials agent is the first living piece of the system, and it does two complementary things. It already runs against ClinicalTrials.gov, PubMed and Europe PMC; in Pharox DX it is integrated as a platform agent, receiving the full clinical map as context.
Layer 1
The agent receives the full case profile from the map —molecular profile, stage, histology, labs and ECOG— and evaluates it against the ~580,000 active studies on ClinicalTrials.gov. Scoring runs in two steps: first deterministic over the protocol's structured criteria, then semantic over the criteria written in natural language.
The result distinguishes three states, always with explicit criteria
Layer 2
No oncologist has time to stay current with every guideline update for every molecular profile among their patients. The agent monitors the sources continuously and, when a relevant update appears for an active case, the alert reaches that specific case —with the map open— rather than arriving as a generic newsletter. The physician reviews it and decides whether the plan changes.
That contextualization is what makes the information actionable.
The copilot presents it to the oncologist as a decision surface. The assistant translates it into the patient's language. One model of the case, two audiences that need different things from it.
Clinician copilot
A dashboard shows data; the copilot presents a reasoned graph. The oncologist sees the state of every clinical area, navigates the relationships between nodes to understand the context of each data point, and asks in natural language —“what therapeutic options are available with the current molecular profile?”— over the data in the map.
Patient assistant
“There's a HER2 test in progress, and your doctor already knows” is more useful to María than “HER2 IHC 2+ requires FISH confirmation.” The assistant translates clinical complexity into the patient's register without losing precision underneath: what they have, what it means, what is in progress and what comes next.
The limit
It never assesses severity, never classifies a symptom as expected or unexpected, never offers clinical reassurance. The reason is design, not tone: fever in a neutropenic patient is a genuine oncologic emergency, and a system that says “that's expected, don't worry” can delay a visit that cannot be delayed. For any reported symptom, the assistant records it in the map with date and time and refers the patient to their care team. The limit is absolute, and it lives in the code before it lives in the policy.
Competing with the global leaders on their axis —capital, time and geography— is unviable, and unnecessary. The real window is in what they do not do: come down to Latin America with a product adapted to the fragmentation of the local health system, to low reimbursement, to heterogeneous regulation, to the language and to the scarcity of oncologists. That combination, a barrier to entry for the incumbents, is our natural operating space.
Legitimacy, ethical consent and a direct channel to the patient. Institutional trust in Latin American healthcare cannot be bought: it is built over years.
Proven strength in ingestion, governance and clinical data graphs — which reduces execution risk in precisely the most critical phase of the project.
Access to Latin America's precision oncology ecosystem. In a project whose value proposition is coordinating it, that access is a product requirement.
Knowledge of the medicine, the regulation, the language and the local codes of Latin America. It cannot be bought or replicated quickly from outside.
Already proven against ClinicalTrials.gov. A concrete starting point that shortens the time to first delivered value.
The combination —geography, institution, engineering, network and technical starting point— is the real moat. Not the most sophisticated algorithm, but the configuration that makes the project viable in the specific context where it operates.
We are in conversation with institutions and clinical teams, with players in the precision oncology ecosystem, and with those evaluating whether to back the project. If any of this resembles what you are looking for, write to us.
See the copilot over a case from your service and understand what it takes to ingest the data you already hold.
Integrate with the map to receive qualified patients, at the right moment and with full clinical context.
Go through the phasing, the project's assets and the regional coordination strategy in detail.