Sanjeevani

Every blood donation saves a life, yet the process of finding a suitable donor during an emergency remains surprisingly manual. Blood banks and volunteer organizations still depend on spreadsheets, phone calls, WhatsApp groups, and human memory to coordinate donations. When experienced volunteers become unavailable, valuable knowledge—such as who is reliable, who prefers evening calls, or who is temporarily unavailable—is often lost. For patients with conditions like thalassemia, who require regular blood transfusions throughout their lives, these delays can be life-threatening. We realized that the problem wasn't a lack of donors—it was the lack of an intelligent system that could remember relationships, learn from every interaction, and coordinate autonomously. That realization inspired SanjeevaniX. What We Built SanjeevaniX is an autonomous blood coordination intelligence platform that combines artificial intelligence, semantic search, persistent memory, multilingual voice agents, and workflow automation to transform emergency blood coordination. Unlike traditional blood management systems that simply store donor records, SanjeevaniX continuously learns from every interaction. The platform can: Receive emergency blood requests instantly through webhooks Verify donor eligibility using strict 90-day medical donation rules Perform hybrid donor matching using BM25 and vector search Retrieve personalized donor context using Retrieval-Augmented Generation (RAG) Contact donors through multilingual AI voice calls powered by Vapi Send WhatsApp location details automatically after donor confirmation Learn from every conversation to improve future matching and outreach Instead of automating individual tasks, we built a system that continuously improves itself over time. How We Built It Our architecture is designed around autonomous AI workflows. Intelligent Workflow Orchestration We built the entire coordination engine using n8n, allowing every stage—from emergency intake to donor confirmation—to operate automatically. Hybrid AI Search To maximize matching accuracy, we combined: BM25 for precise blood group filtering Pinecone Vector Search for semantic similarity and behavioral matching This hybrid approach ensures both medical correctness and intelligent donor prioritization. Persistent AI Memory Every donor interaction is transformed into embeddings and stored inside Pinecone. Instead of remembering only donor details, the system remembers: Previous conversations Language preference Donation history Availability Preferred contact time Reliability Trust score Fatigue score This creates a long-term memory layer that grows smarter after every interaction. Voice AI Accessibility was one of our biggest priorities. Rather than requiring users to install another mobile application, we integrated Vapi Voice AI to communicate naturally through phone calls in regional languages. This allows SanjeevaniX to reach: Rural populations Elderly donors Non-English speakers Users without smartphones Voice became our inclusion strategy—not just another feature. Scalable Infrastructure Our prototype is orchestrated using n8n, while the production architecture is designed around: AWS Lambda Amazon EventBridge Step Functions API Gateway Amazon SQS PostgreSQL with pgvector Pinecone This enables near-zero idle infrastructure cost while supporting national-scale deployment. Challenges We Faced Preserving Human Memory One of the biggest challenges wasn't finding donors—it was preserving volunteer knowledge. Experienced coordinators naturally remember: Who responds quickly Who recently donated Who prefers phone calls Who is unavailable Which communication style works best Capturing this human intuition and converting it into persistent AI memory required redesigning the coordination process from scratch. Medical Safety Healthcare applications cannot depend on probabilistic AI for critical decisions. To eliminate hallucination risks, we implemented: Rule-based blood compatibility Hard-coded eligibility checks 90-day donation cooldown enforcement Artificial intelligence only assists with coordination—not medical decisions. Intelligent Matching Blood group compatibility alone is insufficient. Multiple donors may qualify medically, but not everyone is equally likely to respond. We developed a multi-factor scoring model that considers: Blood compatibility Geographic proximity Historical reliability Donor fatigue Predicted acceptance probability This dramatically improves successful donor outreach. Building AI That Learns Most automation systems repeat the same workflow forever. We wanted SanjeevaniX to improve continuously. Every acceptance, rejection, missed call, or successful donation becomes new knowledge that influences future decisions. This self-learning capability became one of the most technically challenging—and rewarding—parts of the project. What We Learned Building SanjeevaniX taught us that AI delivers its greatest value not by replacing people, but by preserving institutional knowledge. We learned how to combine: Retrieval-Augmented Generation (RAG) Vector databases Hybrid semantic search AI voice agents Workflow orchestration Memory-driven personalization into a single autonomous healthcare coordination platform. Most importantly, we realized that the future of AI lies in systems that remember, adapt, and continuously learn rather than simply automate repetitive tasks. Impact SanjeevaniX transforms blood coordination from a reactive, volunteer-dependent process into an intelligent, self-improving network. By reducing response times, preventing donor fatigue, preserving organizational knowledge, and enabling multilingual voice-based communication, the platform helps hospitals, blood banks, and volunteer organizations coordinate life-saving blood donations faster and more efficiently. Our long-term vision is to evolve SanjeevaniX into a nationwide healthcare intelligence layer capable of supporting blood donation, organ donation, emergency healthcare coordination, and other critical public health services across India. Blood donation isn't a data problem. It's a relationship problem. SanjeevaniX is the intelligence layer that remembers those relationships—so no life is lost because the right donor couldn't be found in time.