ChartLink: Intelligent Patient Search with Band.AI
US healthcare has no national patient ID, so a person's records scatter across hospitals, clinics, pharmacies, billing, parking, and patient apps that don't talk to each other. The usual fix — consolidating everything into one database — is expensive, perpetual, and a privacy liability. And a single typo or name change makes a traditional lookup miss entirely. ChartLink takes the opposite approach: don't move the data, connect it. It puts one specialist AI agent on each system, and a Registrar agent coordinates them through Band in synchronized rounds — fanning out a search, gathering what each system finds, corroborating the candidate that agrees across systems, and broadcasting every new clue so the others can confirm. Records are reasoned together, not pattern-matched, so nicknames, typos, accents, and name changes don't break the match. Nothing is copied, stored, or centralized — and every cross-system access is a Band transcript inside a private, per-patient room, giving compliance a built-in audit trail.
Category tags:"I liked this project because it addresses a real healthcare data-engineering problem: finding the correct patient record across disconnected systems without requiring a large central data-consolidation project first. The strongest part is the concrete federated matching workflow. ChartLink places specialized agents alongside existing systems such as pharmacy, clinic, billing, parking, mobile application, and hospital data. The Registrar acts as the coordinator and arbiter. It takes incomplete information from a user, sends relevant clues to the system agents, compares the returned findings, and then uses newly confirmed details to run another search round. The Amelia Smith scenario makes the value very clear. The initial clues are incomplete: a name, a card ending, and a Toyota key. Instead of failing because no single system has all details, the Registrar combines partial matches from different systems. That then helps reveal additional records such as medication, allergy information, blood type, DNR status, and recent visit history. This is a strong example of iterative evidence-based matching rather than a simple database lookup. I also liked that the project does not treat all matching as exact matching. Real healthcare data contains spelling differences, name changes, incomplete records, and human data-entry errors. The idea of specialized agents reasoning over partial information and returning only relevant evidence is a meaningful use of multi-agent coordination. The Registrar design is particularly good. It acts as the user-facing entry point, coordinates the search, arbitrates conflicting or partial records, and creates a private room for each patient search. The presentation also explains the architecture well, especially the visual sequence showing the Registrar sending clues, receiving partial matches, enriching the search context, and repeating until no new useful evidence is found. The business value is significant. Correct patient matching can reduce duplicate records, billing and claim errors, delays in care, and safety risks related to allergies or medication interactions. The project also presents a realistic alternative to expensive long-term consolidation programs by connecting systems on demand. For future improvement, I would like to see stronger production controls around identity confidence, false-match prevention, consent, access authorization, and clinical review. In healthcare, a wrong positive match can be more dangerous than a missed match, so the final system should show confidence scores, source-system evidence, match rationale, and a clear human verification step before critical information is used in care decisions. I would also clarify the phrase “the data never moves.” The architecture appears to avoid permanent central consolidation, which is valuable, but agents still exchange limited record details for comparison. In production, this should be governed by minimum-necessary data sharing, encryption, audit logging, HIPAA-aligned controls, and role-based access. Overall, this is a highly polished and technically strong multi-agent healthcare interoperability project. It uses Band as a real coordination layer, demonstrates an easy-to-understand iterative matching workflow, and solves a meaningful operational and patient-safety problem. "
Amit Singh
AVP
For Innovators & Creators