Upload Dashboard
Chat Interface
Architecture

🩺 CuraBot — Intelligent Research Assistant for Doctors

Your AI-powered clinical research agent with secure data handling, redis-backed personalization, and automated medical literature discovery.

🚑 💡 Problem

Doctors lose valuable time searching through scattered medical literature, clinical trials, and symptom-specific research. Even with AI, privacy and data accuracy remain major concerns — PHI leakage, hallucinations, and irrelevant search results slow down critical decisions.

🚀 ✨ What We Built

CuraBot is an AI-powered research assistant for doctors.

A doctor uploads the patient's details → sensitive data is immediately redacted → CuraBot performs deep medical research → indexes everything in Redis → the doctor gets actionable, evidence-backed insights through an interactive chat.

The system delivers:

Automated symptom-based medical research
Redis-powered semantic search + personalization
Skyflow-protected patient data with instant PHI masking
Doctor-facing chat interface to interpret research clearly
Postman Flow agent orchestration for rapid prototyping

🔧 Tech Stack

Postman Flows (agent workflow automation)
Redis Vector Library (RedisVL) (semantic retrieval + personalization memory)
Skyflow Data Privacy Vault (PHI redaction + tokenization)
Typescript + Node backend
Parallel Web Systems (high-speed research + content extraction)

🤝 Sponsor Integrations (Why & How We Used Each)

🧡 Redis — Why We Chose It & How It Powers CuraBot’s Intelligence

Why Redis?

CuraBot needed a system that could:

retrieve medical evidence instantly
run semantic search on embeddings
personalize results based on real doctor behavior
adapt across repeated patient cases
stay fast even as the dataset grows

Redis + Redis Vector Library (RedisVL) gave us exactly that: millisecond retrieval, vector similarity search, and stateful personalization — all in one place.

How We Used Redis Inside CuraBot

We store vector embeddings for:

peer-reviewed research articles
clinical trials
medical reviews

RedisVL handles top-k semantic search, letting CuraBot surface the most meaningful evidence for any patient problem—instantly and at scale.

We also track lightweight interaction data in Redis:

which articles the doctor opened
which sources they preferred (journals, trials, reviews)
categories they repeatedly interact with
patterns across similar patient cases

This gives CuraBot a continuously updated picture of what each doctor finds useful.

🧠 Redis Gives CuraBot a Sense of What Matters to Each Doctor

Every interaction — every click, every ignored link, every repeated case — becomes a subtle learning signal stored in Redis.

Over time, CuraBot begins to understand:

preferred evidence types
high-value sources
search patterns across specialties
what answered previous cases effectively

Redis turns these signals into a foundation for smart, adaptive recommendations.

⚡ Redis Lets CuraBot Re-Rank Results in Real Time

Using the data stored in Redis, CuraBot adjusts how results are ordered:

frequently opened sources rank higher
irrelevant ones drop
certain journals get boosted for certain specialists
clinical trials surface when they historically helped
semantic matches from similar past cases guide new searches

This makes each search feel more aligned with how the doctor actually works.

🌱 Redis Makes CuraBot Grow With Every Use

Each query enriches Redis. Each interaction refines future relevance. Each repeated case strengthens embeddings. Each update makes recommendations feel more “right.”

CuraBot doesn’t just fetch evidence — it evolves into a faster, more intuitive assistant for every doctor who uses it.

Redis is the engine behind that evolution.

🔒 Skyflow — Secure Data Masking & Privacy Layer

Why Skyflow?

We needed a robust privacy solution to handle patient uploads securely and avoid accidental PHI exposure. Skyflow's Zero-Trust Data Privacy Vault provided tokenization that goes far beyond simple regex masking.

How We Used It:

Immediately redacted sensitive information right after the upload
Masked: names, DOB, contact details, identifiers, and other PHI
Ensured only de-identified symptom text enters our research + embedding pipeline
Guaranteed all stored data & vector embeddings remain PHI-free