MedSim AI
By
@Leogenie
- 1 Devlogs
- 1 Total hours
An advanced multi-modal medical simulation platform coded locally in Acode and compiled via Termux on mobile.
Ship #1
Changes requested
I built MedSim AI—a finalized, high-fidelity clinical triage and medical simulation platform.
As a 16-year-old student balancing intense academic Higher exams with absolutely zero budget, this project was born entirely out of resource constraints. Lacking access to a traditional laptop or PC, I spent over 100+ hours of intense technical work engineering this entire production stack natively from my phone. While I used AI to generate the code blocks, the heavy lifting was entirely my work: I architected the application flow, authored the complex system prompts, structured the state logic, and manually integrated every piece inside Acode. I hosted a local container runtime using Termux (npm run dev) and handled version control through custom shell aliases optimized to bypass mobile soft-keyboard bottlenecks.
Because I didn’t have a desktop compiler to catch bugs, I had to innovate my own development pipeline to keep the files stable. I built a strict Multi-Model LLM Consensus validation loop: harvesting raw feature logic from a custom Gemini Coding Assistant Gem, pushing the output to ChatGPT for static error diagnostics, running self-correction cycles back into Gemini, and cross-validating the live speech-routing of my Voice Call Mode across Gemini, ChatGPT, Copilot, and Grok simultaneously before doing the manual merges myself.I am incredibly proud of how robust the final architecture turned out.
The live app features a fully state-tracked Objective Physical Assessment lab (vitals, heart rate, HEENT) that reacts to randomized patient profiles and conditional behavioral matrices (anxiety, pain index, fatigue). Everything terminates into an elite, multi-vector diagnostic analytics dashboard that catches clinical red flags and maps performance data over time.I poured my heart into proving that with a resourceful AI orchestration pipeline and pure determination, you don’t need expensive developer rigs or university funding to ship production-grade software. The platform is finalized, responsive, and completely live at medicsim.vercel.app.
🚀 MedSim AI — Final Product Live!.
This devlog captures the absolute completion of the platform. Check out the Trailer here : https://youtu.be/EWaBFTak7wg?is=Sd0Ufs2qORn--ROW
100+ hours went into engineering this setup solo on mobile while balancing Higher exams. Built natively using Acode for source file edits, hosted via local Termux Linux containers (npm run dev), and deployed to Vercel via custom shell terminal aliases..
🧬 THE MULTI-MODEL AI CONSENSUS PIPELINE:Because I lacked a desktop compiler, I engineered a strict cross-model validation loop to guarantee production-ready code blocks
🛠️ Targeted Generation: Core layout architecture was generated using a custom Gemini Gem tuned as a specialized software assistant.
🔍 Error Auditing: Code was run through ChatGPT to systematically catch hidden compile and runtime bugs.
🔄 Self-Correction: Discovered errors were fed straight back to the Gemini Gem to execute self-correction cycles.
🧠 Feature Cross-Validation: High-complexity features—like the real-time Voice Call Mode—were processed through a parallel chain of Gemini, ChatGPT, Copilot, and Grok to eliminate API mismatches before being merged into the local build
🔥 COMPLETED PLATFORM MODULES SHOWN:.- 🏥 10x Accelerated Interactive Lab (/labs): State-tracked Objective Physical Assessments (HEENT, Vitals, Heart Rate) reacting to randomized demographics and behavioral matrices (Anxious, Tired, Quiet, Shy, Talkative).
🎙️ Live Voice Call Mode: Text-to-voice and speech-to-text routing for realistic verbal patient consultation.
📊 Top-Tier Diagnostic Analytics Dashboard: Compiles historical telemetry logs, localized medical specialty tracking (ENT, Dermatological), critical red-flag catchers, and automated notification triggers identifying low-performing clinical sectors.
