🤖 AI Features

Overview

HealthCIP integrates AI at two key points to reduce doctors' administrative burden while maintaining clinical accuracy:

1. AI Treatment Plan Generator — Automatically generates treatment plans from diagnosis and symptoms, grounded in your actual pharmacy inventory
2. AI Chat Assistant — A conversational interface for clinical queries with contextual memory

AI Treatment Plan Generator

How It Works

The AI Treatment Plan Generator uses a two-stage pipeline that's both powerful and cost-efficient:

Stage 1: LLM Suggestion (Cheap)

The AI receives only the patient's diagnosis and symptoms — no master data lists, no large context. It returns a conceptual treatment plan using generic medicine names (e.g., "Paracetamol 500mg tablet"). This prompt is tiny (~100 tokens), keeping API costs near zero.

Stage 2: BM25 Grounding (Zero Tokens)

The generic suggestions from Stage 1 are matched against your actual master data using BM25 text search — the same algorithm used in search engines. This maps generic medicine names to real products in your pharmacy, complete with their actual database IDs. This stage uses zero LLM tokens.

How to Use

1. Start a consultation from the Doctor's appointment list
2. Record symptoms from the symptom master list
3. Enter the diagnosis/inference in the clinical notes
4. Click the AI Treatment Plan button
5. Optionally add Doctor's Notes/Preferences — the AI will prioritize these (e.g., "Prefer generic brands" or "Avoid NSAIDs")
6. The AI returns a complete treatment plan with medicines, dosages, intervals, and durations
7. Review the plan — each item is already linked to your master data
8. Edit, add, or remove items as needed
9. Save the treatment plan to the patient's record

Sample AI Treatment Plan Output

After entering "Acute sinusitis with fever and headache" as the diagnosis, the AI might return:

Medicine	Dosage	Interval	Duration	Timing
Amoxicillin 500mg capsule	500mg	1-0-1	7 Days	AFTER_MEAL
Paracetamol 500mg tablet	500mg	1-1-1	3 Days	AFTER_MEAL
Cetirizine 10mg tablet	10mg	0-0-1	7 Days	BEFORE_MEAL

Each medicine is already matched to your actual inventory — the doctor just reviews and approves.

Technical Details

The AI service is powered by Spring AI and supports multiple LLM providers:

• Default model: Deepseek (configurable via llm.model)
• APi URL: Configurable via llm.api-url (defaults to Deepseek API)
• Temperature: 0.2 (low creativity for clinical accuracy)
• Max tokens: 1,500 per treatment plan generation
• BM25 matcher: Zero-token grounding using text similarity

AI Chat Assistant

How to Use

1. Log in as a Doctor
2. Navigate to Doctor > Chat
3. Type your clinical question in the chat input
4. The AI responds with contextual information
5. The conversation maintains context — you can ask follow-up questions

Chat Features

• Session memory: The AI remembers the conversation within a session
• Context persistence: Chat context is maintained per user session
• Multiple providers: Supports both OpenAI and Deepseek backends

Configuration

To enable AI features, your administrator needs to configure the following in application.properties:

# LLM Provider Configuration
llm.api-key=your-api-key-here
llm.api-url=https://api.deepseek.com/v1/chat/completions
llm.model=deepseek-v4-flash

# OpenAI alternative (uses Spring AI auto-configuration)
# spring.ai.openai.api-key=your-openai-key
# spring.ai.openai.chat.model=gpt-4

The AI also uses Spring AI's ChatClient with MessageChatMemoryAdvisor for chat memory management.

Files Reference

File	Purpose
AiService.java	Interface for AI chat service
OpenAiServiceImpl.java	OpenAI/Spring AI chat implementation
AiTreatmentPlanService.java	Two-stage treatment plan generator (LLM + BM25)
AiInferenceService.java	AI inference for diagnosis support
AiInvestigationService.java	AI-assisted investigation suggestions
Bm25Matcher.java	BM25 text search algorithm for master data grounding