Karim penuel
AI & Data Science DITAi medical scribe
Problem Statement
1.Doctors often misdiagnose due to long typing and loose attentions, leading to incorrect prescriptions and delayed recovery for patients. 2.Hospitals struggle to manage patient flow, resulting in long and unpredictable waiting times for patients seeking timely medical attention.
Proposed Solution
Generative AI / Natural Language Processing (NLP) space, specifically text summarization and structured information extraction. A speech-to-text model transcribes the doctor- patient conversation, including code-switched Swahili/English speech. An NLP model then extracts key clinical elements (symptoms, duration, severity, and the doctor's line of questioning and organizes them into a structured, standardized clinical note (e.g., a SOAP-style summary)), grounded against national treatment guidelines and medicine lists to reduce errors and hallucinated content
Innovation
my project comes to processes swahili conservations detween doctor and their patients
Expected Impact
1. Accurate Pre-Lab Extraction This pillar is the foundation of trust in the system, if extraction is wrong, nothing downstream matters. It is judged by: • Symptom mapping: every symptom, its duration, and severity spoken by the patient is captured correctly (e.g., "tumbo linasokota tangu juzi" → "2 - day history of severe abdominal cramping"). • Doctor's inquiry capture: the doctor's initial clinical focus and questions are captured accurately, without missing context. • Zero hallucination: the AI never invents or assumes symptoms that were not actually discussed a hard requirement, since an invented symptom is more dangerous than a missed one. 2. Operational Speed & Integration Accuracy alone isn't enough if the tool slows the doctor down or breaks their workflow. Speed is what makes adoption realistic in a busy clinic: • Instant description drafts: the structured summary appears on the doctor's screen in under 5 seconds after the consultation ends. • Drastic time savings: doctors spend only 15–30 seconds reviewing and approving the note, down from 3–5 minutes of manual typing.3. High Approval Rate (Low Edit Rate) If doctors have to rewrite most notes, the tool has failed even if it's fast. A high approval rate is the clearest signal that doctors trust the output enough to rely on it daily: • High doctor trust: doctors approve the AI's clinical narrative more than 90% of the time, with little to no manual correction. • Natural workflow: the tool feels like a seamless assistant, letting the doctor maintain eye contact with the patient instead of the screen.