Home Builder Projects Problem Solution How It Works Features Impact My Role Demo
AI Healthcare Product · Solo Founder Build

CarePath Engine

An ambient AI scribe that listens to doctor-patient conversations in Hindi, Marathi & English — and turns them into verification-ready prescriptions, automatically.

▶ Watch Demo Read Case Study
Designed · Built · Shipped Solo
About The Builder

The Person Behind CarePath

Atharv Dhotre
Atharv Sanjivkumar Dhotre ✓
AIML Engineer/ Python Developer/ Cybersecurity
Pandharpur, Maharashtra  —  B.Tech CSE '27

Building a focused career in AI/ML with hands-on experience shipping real systems — from an enterprise RAG knowledge assistant to full-stack SaaS products. Strong Python foundation, Deloitte-certified in cybersecurity, and actively deepening ML expertise through Kaggle, fast.ai, and Andrew Ng's ML Specialization. CarePath Engine is my solo-built attempt at solving a real clinical workflow problem with AI.

7.5
CGPA
4
Shipped Projects
94%
Best Accuracy
2027
Expected Grad
Machine Learning Generative AI LangChain RAG FAISS Python Scikit-learn Hugging Face Cybersecurity Product Thinking
Email Call LinkedIn GitHub
Portfolio

Projects Shipped

A track record of building real, functioning AI and software systems — not just tutorials.

Flagship
CarePath Engine
2026
Speech-to-TextNLPClinical NEROn-Premise AI
  • Ambient AI clinical scribe for Indian OPDs — transcribes code-mixed Hindi/Marathi/English consultations in real time
  • Built a 5-category clinical entity extraction pipeline: complaints, denials, prescriptions, diagnostics, advice
  • Designed phonetic matching against Indian pharmaceutical brand names to prevent dispensing errors
  • Engineered fully on-premise processing to meet India's DPDP data privacy requirements
→ Est. saves 1.5–2 hours per OPD shift per doctor
Enterprise RAG Knowledge Assistant
2026
View Repo →
PythonLangChainFAISSOllamaHugging FaceBAAI/bge-small-en-v1.5
  • End-to-end RAG-based question-answering system for PDF documents using a locally hosted LLM via Ollama
  • Built a semantic retrieval pipeline — document loading, chunking, embedding generation, vector indexing, context-aware generation
  • Implemented BAAI/bge-small-en-v1.5 embeddings with FAISS for efficient semantic search and optimized prompt engineering
  • Runs entirely locally — zero external API calls, full data privacy, zero cloud dependency
→ Reduced manual document search time by ~70%
Sassly — SaaS Subscription Platform
2025
PythonWeb DevelopmentDatabaseDashboard UI
  • Platform helping companies track, manage, and optimize SaaS subscriptions from one centralized dashboard
  • Implemented subscription tracking, renewal alerts, cost analytics, and vendor management
  • Built full product end-to-end — backend logic, database design, frontend dashboard interface
→ Built to reduce SaaS spend wastage for teams
Crop Recommendation System
2025
PythonScikit-learnPandasNumPy
  • ML model recommending the most suitable crop based on soil & climate parameters (N, P, K, temperature, humidity, pH, rainfall)
  • Trained and compared Naive Bayes, Random Forest, and Decision Tree classifiers
  • Performed full ML pipeline: preprocessing, EDA, feature selection, training, evaluation
→ Random Forest achieved ~94% accuracy
The Problem

Indian OPDs Don't Have Time to Type.

Doctors see 40–80+ patients a shift with 2–4 minutes each. Every minute spent writing notes is a minute taken from the patient — and real consultations aren't in clean English.

"दोन दिवस ताप आहे, Paracetamol 650 mg जेवणानंतर दोनदा घ्या"

Global dictation tools break on this kind of Hindi-Marathi-English code-switching. And India's DPDP Act makes cloud-based voice transcription a compliance risk most hospitals won't take.

The Solution

An AI That Listens, Understands, And Drafts.

CarePath runs entirely on-premise. It passively listens during consultation, uses AI to extract structured clinical data, and hands the doctor a pre-filled prescription to approve in one click. No typing. No dictation commands. No cloud.

How It Works

Five AI-Driven Steps

Ambient Listening

Captures natural consultation audio, zero commands needed.

Multilingual Transcription

Tuned for Hindi/Marathi/English code-switching + drug names.

Clinical Extraction

Pulls symptoms, denials, prescriptions, labs & advice.

Doctor Verification

Pre-filled screen — confirm or edit in seconds.

One-Click Output

Print-ready prescription + full audit trail.

What Makes It Different

Built For Real Indian Clinics.

Code-Switch Recognition

Handles the actual way doctors and patients speak — mixed language, mid-sentence.

Phonetic Drug Matching

Calibrated against Indian brand names like Dolo, Pan-D, Augmentin to prevent dangerous misreads.

Safety Gates

Blocks printing if dosage frequency or meal timing is incomplete.

On-Premise Processing

Patient voice data never leaves the clinic — fully DPDP compliant.

5-Category Extraction

Complaints, denials, prescriptions, diagnostics, and care advice — structured automatically.

Full Audit Trail

Transcript, AI draft, and final doctor edits logged for medicolegal traceability.

Impact

What This Means For A Doctor's Day

1.5-2H
Saved Per Shift
40-80+
Patients Per Shift
0
Cloud Exposure
100%
Doctor-Verified
My Role

Solo Founder. Solo Builder.

End-to-End Ownership

See It In Action

Product Demo