PROMPT
BIOTICS.

A unified healthcare AI system combining Vertex AI with USDA FoodData Central. Designed for Type 2 Diabetes, Hypertension, and CKD management.

98.5%

SAFETY RECALL

298k+

USDA ITEMS

01SYSTEM ARCHITECTURE

1. The "Brain": Fine-Tuned LLM

We employ Google's Gemini 2.0 Flash, fine-tuned on a custom corpus of 3,847 clinical-nutritional examples. Unlike generic models, it is optimized for medical terminology and specific dietary interventions.

2. The "Fuel": USDA Integration

The system ingests the USDA FoodData Central database (298,476 items). It retrieves exact micronutrient profiles (Potassium, Sodium, Fiber) rather than hallucinating nutritional values.

3. The "Guardrails": Safety Framework

A deterministic rules engine runs parallel to the AI. It checks for:

Drug-Nutrient Interactions (e.g. Warfarin + Vitamin K)
Contraindications (e.g. CKD + Potassium)
Dosage Hallucinations

FORMULA 1.2: CONFIDENCE SCORING

C_model=

¼ (C_length + C_entropy + C_prob + C_param)

C_coherence=

0.3(Complete) + 0.3(Repetition) + 0.2(Medical)

C_overall=0.78(VALIDATION SET AVG)

02INTERACTIVE SIMULATION

See the Logic
in Real-Time.

The console to the right visualizes the exact pipeline described in Section IV of the research paper. Watch how the system handles a complex query involving Type 2 Diabetes andACE Inhibitors.

Ingestion Latency: 120ms

Inference Time: 850ms

Safety Check: PASS

live_inference_log.sh

03DATASET STATISTICS (SECTION III)

3,847

Training Examples

Clinical Q&A Pairs

298,476

Food Items

USDA Verified

Conditions

Chronic Diseases

Medication Classes

Pharmacological Types