ClarityRx

AI-Powered Indian Prescription Reader & Medicine Price Comparator

Upload a handwritten prescription. Understand every medicine. Find the best price.

Built for · Anakin Build-A-Thon2026

---

The Problem

In India, over 1 billion prescriptions are written every year — most of them by hand, in a mix of English brand names, Hindi instructions, and shorthand that even pharmacists struggle to read. Patients often:

• Cannot read their own prescription
• Don't know what a medicine is for or when to take it
• Overpay because they don't know cheaper alternatives exist
• Miss critical dosage instructions written in regional script

ClarityRx solves all four problems in one upload.

---

What ClarityRx Does

Upload prescription image
         │
         ▼
  Gemini 2.5 Flash OCR  ──────────────────►  Exact transcription
         │                                    Hindi/Devanagari aware
         ▼
  LLM Drug Name Resolver ─────────────────►  "Sinarest" ← "Sinarset"
         │                                    Indian brand knowledge
         ▼
  Fuzzy Medicine Matcher ─────────────────►  Verified against 20k+
         │                                    Indian medicine records
         ▼
  Clinical Plausibility Check ────────────►  Does this make sense
         │                                    for this patient?
         ▼
  Patient-Friendly Summary ───────────────►  "You likely have a
         │                                    respiratory infection…"
         ▼
  Live Price Comparison ──────────────────►  Truemeds ₹27 ← BEST
  (triggered on demand)                      Netmeds  ₹28.5
                                             Tata 1mg ₹29

Live Demo

🔗 clarityrx.tejasdeveloper.xyz

Test credentials — Email: test@test.com · Password: testing

---

Key Features

Feature	Description
🔬 Multi-layer Drug Verification	5-step pipeline from raw OCR to clinically validated match
🗣️ Hindi-Aware OCR	Understands Devanagari script, MAN notation, circled durations
🧠 AI Health Summary	Explains your condition and medicines in plain language
💰 Live Price Comparison	Real-time prices from Truemeds, Netmeds, Tata 1mg — side by side
📊 Confidence Scores	Every drug shows how confident the AI is in its identification
⚡ Parallel Price Fetching	All 3 pharmacies queried simultaneously — results in ~10–20 seconds
🔒 Privacy First	User-scoped data — patients only see their own prescriptions

---

Demo

Upload → Analyse → Compare → Save

Input: Photo of a handwritten prescription (even messy ones)

Output:

• Patient condition banner (Chief Complaint + Age + Weight)
• Medicine catalog — 3-column grid, one card per drug
  • Verified badge + confidence bar
  • Form, frequency, duration chips
  • Plain-English description
  • "Compare Prices" button → live catalog from 3 pharmacies
• AI health summary paragraph at the bottom

---

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                         Django Backend                          │
│                                                                 │
│  views.py                                                       │
│  ├── upload_prescription()   ← runs full pipeline on upload     │
│  ├── view_prescription()     ← renders result page              │
│  └── api_medicine_prices()   ← lazy price fetch on button click │
│                                                                 │
│  Ai_services/                                                   │
│  ├── prescription_pipeline.py   ← orchestrates all AI layers   │
│  ├── ocr_gemini.py              ← Google Gemini 2.5 Flash       │
│  ├── drug_name_resolver.py      ← Groq LLaMA 3.3 70B           │
│  └── groq_simplifier.py         ← Groq LLaMA 3.3 70B           │
│                                                                 │
│  medicine/                                                      │
│  ├── medicine_match.py          ← RapidFuzz local matcher       │
│  ├── medicine_loader.py         ← JSON database loader          │
│  └── indian_medicine_data.json  ← 20k+ Indian medicine records  │
│                                                                 │
│  services/                                                      │
│  └── medicine_price_services.py ← Anakin Wire → 3 pharmacies   │
└─────────────────────────────────────────────────────────────────┘

---

Pipeline Deep Dive

Layer 1 — Gemini OCR

gemini-2.5-flash with extended thinking (thinking_budget=4096) transcribes every character exactly as written. It understands Indian prescription conventions:

• Hindi/Devanagari mixed with English brand names
• 1-0-1 frequency notation (Morning-Afternoon-Night)
• Circled numbers ①⑦⑩ = duration in days
• Common prefixes: Tab., Syr., Cap., Inj.

Layer 2a — LLM Drug Name Resolver

llama-3.3-70b-versatile on Groq takes garbled OCR names and suggests the most likely real Indian brand or generic name, using patient age, weight, and co-medications as context clues.

Layer 2b — Fuzzy Medicine Matcher

rapidfuzz.partial_ratio against a 20k+ local JSON database of Indian medicines. Prefix pre-filtering for speed. Threshold: 75/100 to be marked verified.

Layer 2c — Clinical Plausibility Validator

A second LLM call checks whether each matched drug makes clinical sense for this patient's condition, age, and other medications. Implausible matches are flagged and sent for re-matching.

Layer 3 — Re-match Loop

Drugs flagged as implausible are re-run through the fuzzy matcher using the LLM's suggested correction. Better matches are adopted automatically.

Layer 4 — Patient Descriptions + Health Summary

Single Groq batch call generates one-sentence plain-English descriptions per drug. A separate call synthesises a 3–5 sentence health summary explaining the likely condition.

Price Layer — On-Demand, Parallel

Triggered only when user clicks "Compare Prices". Three Anakin Wire async jobs (tm_search, nm_search, tmg_search) submit simultaneously via ThreadPoolExecutor, results merged and sorted cheapest-first.

---

Tech Stack

Layer	Technology
Web Framework	Django 4.x
OCR	Google Gemini 2.5 Flash
LLM	Groq — LLaMA 3.3 70B Versatile
Fuzzy Matching	RapidFuzz
Medicine Database	Custom Indian Medicine JSON (20k+ records)
Price APIs	Anakin Wire (Truemeds · Netmeds · Tata 1mg)
Frontend	Django Templates · Tailwind CSS CDN · Font Awesome
Database	SQLite (dev) · PostgreSQL (prod)
Language	Python 3.11+

---

🖼️ Screenshots

Desktop	Mobile

Installation

# Clone
git clone https://github.com/Noob-Developer-Real/ClarityRx.git
cd ClarityRx

# Virtual environment
python -m venv .venv
source .venv/bin/activate

# Install
pip install -r requirements.txt

# Configure
cp .env.example .env
# Add your API keys (see Environment Variables below)

# Database
python manage.py makemigrations
python manage.py migrate

# Run
python manage.py runserver

---

Environment Variables

SECRET_KEY=your-django-secret-key
DEBUG=True
ALLOWED_HOSTS=127.0.0.1,localhost

GEMINI_API_KEY=your-gemini-api-key
GROQ_API_TOKEN=your-groq-api-key
ANAKIN_API_KEY=ask_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

# Optional — affects Tata 1mg delivery ETA
MEDICINE_PRICE_CITY=New Delhi

---

Team

Built with ❤️ by Noob Mon — solo project for Anakin Build-A-Thon 2026

---

License

MIT — free to use, modify, and distribute.