LISA (Laboratory Information System Application) chat bot

LISA is a web based Laboratory Information System that handles the whole laboratory workflow, starting from specimen tracking, lab jobs with barcode, recording the examination results (both quantitative and qualitative), until generating the printable medical check up report.

With that much data, the admin (doctor or medical personel) needs a lot of time to open the pages one by one just to know the condition of a patient. Therefore, we built an AI chat bot inside the application, so the admin can just ask in natural language and directly get a fast analysis, for example the chronology of a specimen, the summary of a patient condition from the first until the latest examination, checking the test history, or searching a specimen by barcode.

My role is as an AI engineer, and here is what I have done.

First, the AI chat platform itself, both slicing (changing the image UI in Figma into html, css, and js code) and integration, so the chat can be opened as a floating widget from any page of the application without interrupting the admin work.

Second, the RAG management and its API, where the knowledge base can be managed with CRUD and stored as vector embeddings in PostgreSQL, then retrieved using cosine similarity algorithm and consumed by the LLM as context, so the answer is always based on the real laboratory data.

Third, guardrail (both by code and by LLM), which is very important here because the data is medical data, so the chat bot must not answer outside its scope or leak the data of another patient.

And the last one is the system prompt and the chat history management, so the assistant keeps its persona and still understands the previous context of the conversation.