Pulse

The company appears to be a single-person project (Mohamed Mostafa) building an AI-powered clinical assistant for cardiac arrest teams. The author states this is a hackathon submission for the OpenAI 2026 hackathon, built with Codex and various AI tools including GPT-5.5 and GPT-5.6. It is described as a deterministic system that listens during cardiac arrest and maintains clinical state without using LLMs in the clinical loop.

• The description states Pulse is an AI teammate that listens during cardiac arrest
• It maintains clinical state by processing spoken phrases from the resus room
• Spoken evidence becomes clinical events that advance a deterministic state machine
• The system shows team leaders what's true right now and what's due next
• It handles real-world speech including code-switched Egyptian Arabic
• The system is described as having no LLM in the clinical loop, with LLMs only used for speech recognition
• It includes features like CPR clock, rhythm tracking, shock count, medication timeline, and audit trail
• Built with FastAPI + Pydantic backend, Next.js + TypeScript frontend, pluggable speech recognition

• The description states the author is a cardiologist who was inspired by the cognitive load during cardiac arrest
• The product is positioned as solving the problem of team leaders tracking multiple variables simultaneously
• It claims to be deterministic rather than probabilistic, with no LLM in the clinical loop
• The evolution shows from a general idea (team leader gets tired and loses count) to a specific technical solution (deterministic state machine)
• The author states they wrote a "constitution" for the agent that fixes boundaries it cannot cross
• The positioning is described as solving a real clinical problem rather than being a generic AI tool

• The description states the target customer is cardiac arrest teams in hospitals
• It specifically mentions the "Cairo resus room" and Egyptian Arabic speech patterns
• The author identifies as a cardiologist, suggesting they are targeting fellow medical professionals
• No evidence of specific hospital partnerships or institutional adoption
• Not evidenced: whether this targets emergency medical services, training institutions, or other healthcare settings

• Not evidenced: any business model, pricing structure, or monetization strategy
• The description states this is a hackathon submission with no revenue or customer data
• No evidence of subscription models, licensing fees, or institutional pricing
• No evidence of B2B or healthcare market positioning beyond the clinical use case

• Built with Codex across GPT-5.5 and GPT-5.6 sessions
• Backend: FastAPI + Pydantic, frontend: Next.js + TypeScript
• Uses pluggable speech recognition with offline capability
• Includes deterministic state machines, evidence fusion layer, multilingual normalization
• Has confirmation policy and audio pipeline components
• Features include: CPR clock, rhythm tracking, shock count, medication timeline, audit trail
• The system is described as runnable offline with deterministic fake ASR
• The architecture decision states there's no language model in the clinical loop
• Includes testing with replays of real recorded clinician speech
• Addresses challenges like silence hallucination and echo duplication

• Not evidenced: any traction, revenue, customers, or adoption metrics
• This is described as a hackathon submission (OpenAI 2026)
• No evidence of product-market fit, user feedback, or iterative development beyond the single prototype
• No evidence of clinical validation, regulatory approval, or deployment in real hospitals
• No evidence of team growth, funding rounds, or market expansion plans

• Not evidenced: any competitive landscape or existing solutions in this space
• The description states that current tools record what happened but don't understand what's happening
• No evidence of competitors, market size, or differentiation from existing medical technology
• No evidence of regulatory environment or compliance requirements for similar products
• The author mentions the "code-switched Egyptian Arabic" as a unique challenge, suggesting this may be a niche application

• Single-person development: Only one team member (Mohamed Mostafa) is mentioned
• No clinical validation: No evidence of testing with actual medical professionals or institutions
• Hackathon prototype: This is described as a hackathon submission, not a developed product
• Unverified claims: The description states it's self-reported and unverified
• Regulatory risk: Medical applications require extensive validation that's not evidenced here
• Technical challenges: The system addresses issues like silence hallucination and echo duplication, suggesting complexity in real-world speech processing
• Market risk: No evidence of market demand or institutional adoption beyond the author's personal experience

1. What clinical validation has been performed with actual medical professionals?
2. How does this system handle edge cases not covered in the current implementation?
3. What is the regulatory pathway for medical device deployment?
4. Are there any partnerships or pilot programs with hospitals or medical institutions?
5. How does the system handle different accents, dialects, and speaking patterns beyond Egyptian Arabic?
6. What are the specific technical challenges that remain unresolved?
7. How would this system be integrated into existing hospital workflows?
8. What is the timeline for moving from prototype to clinical deployment?