Heidi scales AI clinical documentation with MongoDB

Australian health technology company Heidi has scaled its AI clinical documentation product to 81 million medical consultations globally after building its data platform on MongoDB Atlas.

Heidi Scribe uses speech-to-text and other AI features to generate clinical notes and related administrative outputs. The company said clinicians can spend up to 40% of working hours on documentation tasks. Heidi said its tools have reduced the burden of that work, and it reported more than 18 million hours returned to frontline clinicians over an 18-month period.

Heidi said Heidi Scribe is used in emergency departments, general practice and specialist clinics. The company said the product now supports more than 2.3 million consultations each week across more than 190 countries.

Data platform

MongoDB Atlas sits at the core of Heidi's platform, according to the companies. Heidi said it needed to consolidate medical data from multiple sources, including forms, referrals and clinicians' notes. It also said it needed to manage evolving data types and new application use cases. Heidi also pointed to security and regulatory requirements tied to processing medical data.

Heidi said it built Heidi Scribe on a JSON document database from the start. It initially used Amazon DocumentDB on AWS, before shifting to MongoDB Atlas as its scale and data requirements changed.

"We couldn't scale without downtime, which was a critical issue: we operate in the world of healthcare where clinicians need seamless access to resources 24/7," said Oscar Lukersmith, Head of Data, Heidi. "Our initial database set-up couldn't accommodate the level of growth that our users needed, it didn't support search and index building functionalities-which are key in AI use cases-and we were experiencing increased latency."

The companies positioned MongoDB's document model as a better fit for handling varied data compared with relational databases. Heidi said the schema flexibility aligned with shifting requirements as it added new features and expanded to new markets.

Search and retrieval

Heidi also highlighted its use of MongoDB Atlas Vector Search. The company said it reduces the need for a separate vector database alongside an operational database. It described retrieval-augmented generation workflows that pull relevant information into AI prompts from data stored in MongoDB Atlas.

Heidi said it has built multiple AI tools on top of the platform. It cited clinical coding as a feature that translates patient records into standardised alphanumeric codes used in health systems for billing, reimbursement and reporting. Heidi also referenced "Ask Heidi", which it described as a tool for administrative work such as collecting patient histories, creating ward round lists and conducting audits.

"MongoDB's document model has been a game-changer for our developers. Its flexibility enables us to quickly adapt our AI applications to new use cases-helping us scale to more than 2 million consultations per week, without downtime or bottlenecks," said Ocha Cakramurti, Senior Software Engineer, Heidi "Since migrating to MongoDB Atlas, we've been able to reduce latency on key APIs by nearly one-third, ensuring seamless experiences for medical professionals in critical environments."

Heidi also described a product called Heidi Vector Scribe. It said the tool converts medical documents into vector embeddings and uses semantic search for information retrieval across large volumes of content. It also referenced Langchain as part of its workflow for extracting and embedding medical terms in Atlas.

Operational scale

MongoDB said many AI projects face operational challenges once systems move beyond experimentation and into production environments. It said healthcare workloads place a particular emphasis on reliability and availability.

"One of the biggest challenges companies face with AI isn't experimentation, it's operating reliably at scale-reliability is particularly critical for industries like healthcare. Many tech teams today are managing fragmented combinations of operational databases, vector stores, and model APIs, which leads to unnecessary complexity, latency, and operational risk," said Simon Eid, Senior Vice President, APAC, MongoDB.

MongoDB also framed Heidi's architecture as an example of combining operational data and AI search tooling in one platform. "Heidi is a great example of how choosing a flexible, multi-cloud data platform with embedded AI capabilities can empower developers to quickly move from prototype to production, and then scale," said Eid.

Heidi also linked its technology choices to recruiting and retention in a competitive market for software engineers. "For tech talent, exciting technology matters-the scalability and efficiency of MongoDB Atlas make it a cornerstone of our success, helping us attract developers who want to spend time transforming healthcare, rather than managing databases," said Cakramurti.

Heidi said it is exploring a broader ecosystem of tools for clinical activities, and it described Heidi Scribe as an "agentic AI layer" in that work.