My CareSathi setup requires Python 3.8+; a Modern browser (Chrome or Edge recommended for voice API support); sufficient local system resources to run AI models.
To set up the project, I clone the MedInsight repository, install its Python dependencies, add the local model files, start the Flask backend, and open the frontend in a browser.
git clone https://github.com/Vaatsal-source/MedInsight.git
cd MedInsight
pip install flask flask-cors llama-cpp-python vosk numpy opencv-python
The required local model files are tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf; vosk-model-small-hi-0.22/. I place them in a models/ directory in the project root.
I start the backend with python server.py. The expected backend port and documented URL are 5005; http://<redacted>:5005.
My setup instructions direct me to open index.html in a browser to access the CareSathi dashboard and patient workflow.
My project structure includes index.html, triage.html, analysis.html, inventory.html, server.py, inventory.json, models/, README.md, and optional assets/.
My typical workflow captures patient details and symptoms, enters vital signs such as temperature, BP, and SpO2, uses voice input, receives AI-assisted triage suggestions, reviews X-ray or skin-lesion scans, updates medicine inventory, and generates PDF referral documents.
CareSathi's intended use settings include community health centers, rural clinics, mobile health outreach programs, nurse-led primary care units, and frontline healthcare facilities with limited digital infrastructure.
CareSathi is a decision-support tool intended to assist trained healthcare personnel; it does not replace clinical expertise, licensed medical judgment, or formal diagnosis by a qualified doctor.
It should not be used as the sole basis for diagnosis or treatment decisions.
Final medical decisions must always be made by qualified medical officers or licensed physicians.
I require human oversight in all clinical decisions when using CareSathi.
My documentation says: No license specified; MIT, Apache 2.0, and GPL suggested as examples. It recommends adding a license file before public release.
My roadmap lists stronger multilingual speech recognition; better diagnostic confidence scoring; expanded image analysis models; improved patient record export; more robust inventory analytics; offline patient referral workflows; enhanced UI accessibility and localization.
For questions, collaboration, and project updates, my documentation directs readers to contact the repository owner or maintainers through the GitHub project page. It reports no clinical performance results or roadmap owners or dates.
I describe CareSathi as an integrated health-tech solution designed to assist healthcare workers in rural or low-resource settings.
MedInsight presents CareSathi as an offline-friendly healthcare support suite for frontline workers. It combines clinical decision support with everyday workflows, aiming to help where internet access and infrastructure are limited. This project is part of my GitHub work, alongside the projects summarized on my linkedin profile.
I maintain the repository on @Vaatsal-source's GitHub, and its listed repository description is HTML.
I record the repository dates as: Started 2026-04, last updated 2026-10.
I describe CareSathi as an AI-powered healthcare platform designed to support frontline medical workers in rural and low-resource settings.
The project identifies limited connectivity, specialist shortages, delayed referrals, manual record-keeping inefficiencies, and poor visibility into essential-medicine inventory as challenges for rural healthcare facilities.
I describe the integrated workflows this way: It combines local AI models for triage, voice-assisted data entry, visual diagnostics, and inventory tracking into a single offline-friendly system.
CareSathi's triage assistant analyzes patient symptoms and vitals such as temperature, SpO2, and blood pressure using a local LLM.
CareSathi supports hands-free data entry using multilingual speech recognition for offline use.
CareSathi performs automated analysis for chest X-rays and skin lesion images using computer vision techniques.
CareSathi tracks and updates medical stock such as paracetamol, ORS, and antibiotics in real time.
The listed supported languages are English, Hindi, Odia, and Urdu.
CareSathi generates patient referral documents as PDF files for transfer to higher-level care centers.
Using local models and offline-capable components is intended to provide fast clinical decision support despite limited connectivity and minimal infrastructure. Multilingual interaction is intended to help rural healthcare workers, and PDF referrals support transfer to higher-level care.
The frontend stack is listed as HTML5.
The frontend documentation also lists CSS3, vanilla JavaScript, responsive UI with localized text rendering, and jsPDF for medical reports.
The backend stack is listed as Python Flask.
The backend and storage notes also list Flask-CORS, local JSON-based storage, and browser LocalStorage for patient queue and session persistence.
The language model is TinyLlama 1.1B (GGUF format) via llama-cpp-python.
The speech-recognition component is Vosk (offline Hindi speech model).
The computer-vision and processing stack lists OpenCV and NumPy.
The described system uses local model-based inference for offline operation. The available project description does not report clinical validation, safety evaluation, performance testing, accuracy figures, or measured clinical outcomes. It also leaves open which models and methods power triage and visual diagnostics, how inventory updates work, what hardware and data-protection measures apply, which speech languages are supported beyond the stated offline Hindi model, and how the MedInsight label relates to the CareSathi name.