Patients and caregivers living with uncertainty between appointments, treatment updates, and care-team contact.
Companion
Baseera Companion
A steady presence between clinical moments.
A bounded, longitudinal support layer designed for the hours, days, and weeks between clinical contact.
Offer calm, contextual support between clinical moments without drifting into diagnosis or false reassurance.
Companion supports orientation and understanding. It always defers clinical judgment back to the care team.
Overview
Each project starts from a visible systems failure, then narrows toward a bounded response that people can actually use.
01
The space between appointments
A cancer diagnosis does not pause between appointments. The uncertainty does not wait.
Clinicians are expert, trusted, and stretched thin. Patients leave consultations with answers — and then, quietly, with more questions.
02
The role
Baseera Companion exists in that space.
It is not a diagnostic tool. It is not a replacement for anyone. It is a bounded, longitudinal support layer designed for the hours, days, and weeks between clinical contact.
03
The boundaries
Companion adapts to treatment stage, persona, and context. A patient mid-treatment has different needs than a caregiver in the early weeks of a loved one’s diagnosis.
Every response is grounded in trusted oncology sources. Every boundary is explicit. When a question exceeds the system’s scope, Companion says so clearly: “Please ask your care team.”
04
The principle
The design principle is restraint — because getting this right means knowing exactly where a support tool ends and clinical care begins.
The direction is a continuous, contextual layer of care that travels with the patient across the treatment journey.
Support Arc
Companion is less about one answer than about staying useful across time. The product has to shift with the patient’s stage, confidence, and level of need.
Meet the patient where they are
The system starts from the current context: newly diagnosed, mid-treatment, caregiver support, or long-interval follow-up.
Respond with bounded support
It helps with orientation, explanation, and preparation using trusted sources, while staying explicit about what it cannot decide.
Carry context forward
The experience becomes more useful when it remembers stage, concerns, and recurring themes across the treatment journey.
Escalate when the boundary is reached
When a question becomes clinical, urgent, or personally specific, Companion routes the user back to the care team rather than improvising.
How It Works
The implementation stays deliberately constrained: governed sources, explicit boundaries, and infrastructure that fits the setting it operates in.
01
Foundation
Companion’s first phase was built on OpenAI language models with Qdrant as the vector database — a RAG architecture where every response is grounded in a curated oncology knowledge base before anything is generated.
Rather than a general-purpose model answering freely, the system retrieves first, then responds — keeping answers traceable, bounded, and clinically appropriate.
02
Behaviour and infrastructure
The system was tuned at the behavioural level, not just the prompt level. Adults in active treatment, paediatric patients, caregivers, and elderly users each receive a meaningfully different experience.
The second phase moves the architecture to Azure within Qatar’s sovereign cloud infrastructure, with tighter governance, persistent longitudinal context, and treatment-stage awareness built into the session layer.