Building Similia: An AI-Powered Homeopathic Assistant

The Beginning: A Mother's Wisdom Meets Technology

For me, it began in the most personal way possible as watching my mother, a dedicated homeopath, spend countless hours managing patient cases, flipping through thick repertory books, and carefully analyzing symptoms to find the perfect remedy.

She would often work late into the night, cross-referencing symptoms across multiple books, taking detailed notes, and trying to remember patterns from hundreds of previous cases. I saw the passion in her work, but also the inefficiency. "There must be a better way," I thought.

That's when the idea for Similia was born.

The Journey: From Concept to Reality

Act 1: Understanding the Domain

The first step was understanding homeopathic practice. I was able to get contexts on her workflow:

• Case Taking: 30-45 minutes per patient, handwritten notes
• Symptom Analysis: Manually categorizing and prioritizing symptoms
• Repertorization: Flipping through multiple books (Kent, Synthesis, Boericke)
• Remedy Selection: Cross-referencing, comparing, analyzing
• Follow-ups: Tracking patient progress across weeks and months

Key Insight: The process wasn't broken but it was just waiting to be enhanced by technology.

Act 2: Technical Architecture Design

I knew this wouldn't be a simple CRUD app. It needed intelligence, context awareness, and deep integration with homeopathic knowledge bases.

Act 3: The ReAct AI System

The heart of Similia is its AI system but not just a chatbot, but an intelligent reasoning engine that mimics how homeopaths think.

The ReAct Pattern (Reasoning + Acting):

1. Perception: Understands patient context
2. Reasoning: Decides what information is needed
3. Action: Executes search, asks questions, or provides analysis
4. Execution: Formats and delivers results
5. Memory: Maintains conversation context
6. Evaluation: Learns from interactions

Mother's Reaction: "It asks the same questions I would ask. It's like it knows what I'm thinking."

Act 4: Building the User Experience

Key Features She Loved:

1. Dynamic Multiple Choice: "It guides me through case taking like having a conversation"
2. Automatic Repertorization: "Saves me 20 minutes per case so I can see more patients now"
3. Case History: "Everything is right there, organized so no more lost notes"
4. OCR for Test Results: "Just snap a photo, it reads it! My patients love this"
5. Smart Symptom Analysis: "It picks up on patterns I might have missed"

The Moment That Changed Everything

The first time I saw my mother use Similia with a real patient was unforgettable.

The Scene: Her home clinic, patient sitting across from her, laptop open. My mother, usually surrounded by towers of repertory books, had just her laptop and a calm confidence.

What Happened:

1. She opened Similia, started a new case
2. Patient: "Chronic headache with nausea for 6 months. Nothing helps."
3. AI generated contextual follow-up questions about headache characteristics
4. My mother clicked through the multiple-choice options, occasionally adding her own notes
5. After 8 minutes of guided conversation, she clicked "Analyze & Suggest Remedies"
6. Within 3 seconds, ranked remedy suggestions appeared with detailed justifications
7. She studied the top 3 remedies, cross-referenced with the patient's constitution
8. Selected "Nux Vomica 30C" with confidence
9. Case saved automatically with complete symptom timeline and reasoning

Time Taken: 12 minutes (previously 35-40 minutes)

The Patient's Response: "Doctor, you've upgraded! This is so professional. And you're so much calmer now."

My mother smiled 😊

Her Words Later That Evening: "You know what the best part is? I'm not tired after 6 consultations anymore. I can focus on the patient, on healing, on listening. The AI handles the tedious parts. It's like having a junior homeopath who never gets tired, never makes calculation errors, and never forgets a rubric."

This was about giving my mother more time, more energy, more joy in her work.

The Technical Wins

1. Performance Optimization

Total Response Time: < 1 second for most queries

3. AI Evaluation Results

After rigorous testing:

• Rubric Matching Accuracy: 91% (target: >85%) ✅
• Remedy Relevance: 88% ✅
• Clinical Workflow Completion: 94% ✅
• User Satisfaction: 97% (based on my mother's feedback 😊)

The Impact: Real-World Usage

After 3 months of my mother using Similia:

Quantitative Metrics

Metric	Before Similia	After Similia	Improvement
Avg Case Taking Time	35-40 min	12-15 min	66% faster
Cases per Day	6-8	12-15	87% more
Repertorization Time	15-20 min	2-3 min	88% faster
Patient Satisfaction	4.2/5	4.8/5	14% higher
Follow-up Tracking	Manual, scattered	Automated, organized	100% improvement

Qualitative Feedback

My Mother's Journal Entry (translated):

"I've been practicing homeopathy for 25 years. I never thought technology could understand the nuances of case taking. Similia proves me wrong. It doesn't replace my judgment but it enhances it. It handles the tedious parts so I can focus on healing. My son has built something truly special."

Her Favorite Features:

1. AI Chat Assistant: "Feels like a conversation, not a form"
2. Automatic History: "Never lose patient details and it remembers better than I do"
3. Quick Repertorization: "20 minutes of book-flipping → 2 minutes of AI analysis"
4. OCR Test Uploads: "Patients send photos, AI reads lab results and it's like magic!"
5. Professional Dashboard: "Beautiful and useful which makes me look modern to young patients"

Unexpected Benefits:

• Confidence Boost: "I second-guess myself less now. The AI validates my thinking."
• Teaching Tool: "I show medical students how I diagnose. The AI explanations help them learn."
• Patient Trust: "Patients see me using technology. They trust the treatment more."
• Work-Life Balance: "I finish work by 6 PM now. Time for family. Time for myself."

The Technical Deep Dive:

Architecture Highlights

ReAct AI System Flow

Vector Search Implementation

Performance: Sub-second semantic search across 82,000+ homeopathic rubrics.

Lessons Learned

1. User Feedback is Gold

Every feature was shaped by my mother's feedback:

• ❌ Initial design: Complex admin panel → ✅ Simple dashboard
• ❌ Rigid questionnaire → ✅ Dynamic AI questions
• ❌ Manual remedy entry → ✅ One-click selection with potency

2. AI Needs Constraints and Personality

GPT is powerful but needs guidance and character:

• ✅ Structured outputs for consistency
• ✅ ReAct pattern for logical reasoning
• ✅ Context limits to prevent hallucinations
• ✅ Evaluation metrics for quality control

3. Documentation Is Your Future Self's Best Friend

Creating comprehensive docs helped me:

• Clarify architecture decisions before coding
• Onboard future contributors (even if they're 6-months-older me)
• Remember why I made choices during 2 AM debugging sessions
• Debug faster by having a map of the codebase

4. Build for the 80-Year-Old and the 18-Year-Old

My mother isn't tech-savvy. If she can use it, anyone can.

• Large fonts: Easy to read
• Clear labels: No jargon
• Forgiving UI: Undo is everywhere
• Helpful errors: "Oops! Let's try that again" not "Error 500"
• Instant feedback: Loading states, success animations, gentle failures

Design Principle: If my mother needs to call me to use a feature, that feature is broken.

The Road Ahead

Community Building

Planning to open-source core components:

• ReAct AI engine
• Homeopathic knowledge base
• Vector search utilities

Goal: Build a community of developer-homeopaths

Conclusion:

Building Similia taught me that:

1. Solving real problems for real people you care about
2. Listening deeply to user needs, especially when the user is your mom
3. Iterating relentlessly based on feedback, even when it hurts your ego
4. Caring deeply about the outcome beyond technical excellency or résumé lines

The best part? She tells every homeopath she meets about it. She's my best marketer, my toughest critic, and my proudest user.

The developer in me learned technical skills: AI integration, vector databases, user-centric development.

My mother uses Similia every day.

That's my biggest achievement.

Not the technology. Not the AI. Not the code.

The impact on one person who matters most.