Mrunmayi Bhavsar

image A Multi-Agent AI System for Predictive ICU and Oxygen Demand Management in Emerging Cities MedLyf is an AI-powered predictive healthcare platform designed to forecast ICU and oxygen demand across hospitals in Tier-2 and Tier-3 cities of India. The system addresses the critical oxygen shortage challenges exposed during COVID-19 by transforming reactive crisis management into proactive predictive healthcare. By forecasting demand up to 7 days in advance, optimizing resource allocation and coordinating multilingual logistics, MedLyf ensures healthcare systems can anticipate and respond to surges before they become crises. The Problem During COVID-19, Tier-2 and Tier-3 Indian cities faced catastrophic oxygen shortages due to reactive crisis management instead of predictive planning. Preventable deaths occurred when resources remained unused at some hospitals while others faced critical shortages. The Solution MedLyf forecasts ICU bed requirements and oxygen demand up to 7 days in advance using autonomous AI agents, enabling proactive resource allocation before crises develop. System Architecture: Six Autonomous Agents Built with CrewAI Multi-Agent Framework image Data Ingestion Agent • Purpose: Monitors hospital systems, vendor APIs, weather data, and disease reports; detects data drift and triggers retraining. • Tech Stack: Python workers with PostgreSQL for storage, scipy for statistical drift detection. Forecasting Agent • Purpose: Predicts ICU occupancy and oxygen demand (liters/minute) at hospital, district, and city levels 5-7 days ahead. • Tech Stack: Prophet for time-series forecasting, scikit-learn for baseline models, TimescaleDB for time-series data. Resource Optimization Agent • Purpose: Calculates optimal patient transfers, staff reallocation, and oxygen redistribution across hospital networks. • Tech Stack: scipy.optimize for linear programming, OR-Tools for constraint solving. Logistics Coordination Agent • Purpose: Schedules oxygen deliveries, tracks GPS locations, reroutes ambulances dynamically, and issues supply chain alerts. • Tech Stack: Node.js with Express for vendor API integrations, Mapbox for routing, WebSocket connections for real-time GPS tracking. Communication Agent • Purpose: Generates multilingual advisories (Hindi, Marathi, English) and sends them via WhatsApp, SMS, and email when thresholds are crossed. • Tech Stack: Node.js with Twilio API for SMS, WhatsApp Business API, Jinja2 template engine for message generation. Feedback & Learning Agent • Purpose: Compares predictions vs. actuals, calculates error metrics (RMSE, MAE, MAPE), and triggers retraining cycles to maintain confidence scores. • Tech Stack: scikit-learn for metrics calculation, PostgreSQL for logging, automated retraining pipeline triggers. Real-World Impact • Proactive Planning: 5-7 day advance warnings enable preemptive resource reallocation before capacity breaches. • 70% Less Manual Work: Automated logistics coordination reduces phone calls and spreadsheet management. • Multilingual Access: Hindi, Marathi, and English advisories reach rural frontline workers. • Continuous Improvement: Self-learning system improves accuracy with every forecast cycle. • Scalable Architecture: Pilots with 10-50 hospitals, scales to state-level deployments. image Bottom Line MedLyf uses CrewAI's role-based multi-agent framework to autonomously orchestrate demand forecasting, resource optimization, logistics dispatch, and stakeholder communication. The system prevents oxygen crises through predictive planning rather than reactive firefighting, ensuring optimal resource distribution across hospital networks before emergencies develop.

TapOut The Problem Paying someone in crypto shouldn't require a 5-minute conversation: "What chain are you on?" "Do you have USDC or ETH?" "Send me your wallet address" Wrong network → stuck funds → frustration Token and chain fragmentation broke casual payments. What TapOut Does | Before | With TapOut | |--------|-------------| | Share wallet address | Share phone number or ENS | | Agree on chain & token | AI auto-negotiates from ENS prefs | | Manual bridge + swap | AI routes & executes automatically | | 3-5 apps, 5 minutes | 1 tap, ~30 seconds | Use Cases In-Person (NFC) Split dinner bill Pay a friend back Community meetups & events Remote (ENS) Send money across cities/countries Pay freelancers globally Family remittances What Makes It Better | Aspect | Improvement | |--------|-------------| | Easier | Phone number = wallet. No addresses. | | Faster | AI handles routing. One confirmation. | | Safer | Forwarding contracts hide main wallet. | | Cheaper | AI finds lowest gas + best route. | Tagline Crypto payments, finally simple.