docqueue

the shape of the problem

Hospital triage has two failure modes that look the same from the outside. Either the queue is ordered badly, so the wrong patient is seen first, or nobody knows where the ambulance is, so the clock starts late. Fixing one without the other does not move the number that matters.

DocQueue attacks both: a classifier that reads the symptom description, and a scheduler that orders the queue from it.

what i built

• A two-stage intelligent triage engine, extending an IEEE-published dynamic decision-tree and round-robin scheduling algorithm with a fine-tuned Mistral-7B symptom classifier — reaching 92% diagnostic triage accuracy.
• A real-time bi-directional tracking module over Socket.IO, reducing ambulance dispatch time by 45% and triage time by 73% (15 minutes to 4).
• The end-to-end production system: a hospital staff web console and a React Native patient app, backed by a Node.js and AWS Lambda microservice architecture over MongoDB.

the two-stage decision

The published decision-tree algorithm is deterministic and auditable, which is what a hospital needs. A language model is good at reading a messy free-text symptom description, which is what a patient actually gives you. Stacking them keeps the model at the edge of the system — it classifies the input, the tree still makes the call — rather than putting a probabilistic component in the position of deciding who is seen first.

outcome

A shipped end-to-end system: staff console, patient app, and the triage engine behind both.