Hi, I'm Harsh
Building intelligent healthcare systems at Smart Government Technology LLC. Specializing in Machine Learning, NLP, Generative AI & production-grade HIPAA-compliant software.
About Me
AI Engineer passionate about building intelligent systems that make a real impact
I'm an AI/ML Engineer with hands-on experience building and deploying end-to-end Machine Learning pipelines for Computer Vision, NLP, and Generative AI applications — from academic research to production systems used by real users.
I currently work remotely at Smart Government Technology LLC (formerly Pulscribe.ai) in Virginia, USA, building an AI-powered hospital management platform end-to-end — from FastAPI backend and system design to AI model research and HIPAA/PHI-compliant deployment.
Across projects, I move between research — fine-tuning, quantization, model evaluation — and production engineering — APIs, databases, Docker/Kubernetes, monitoring. I like building things that hold up under real usage, not just demos.
AI Engineer
Currently at Smart Government Technology LLC (Virginia, USA), building a comprehensive AI-powered hospital management platform.
Full-Stack AI
From model research & fine-tuning to FastAPI backends, Kubernetes deployments, and HIPAA-compliant system design.
B.Tech in AI/ML
Pursuing B.Tech in AI & ML at CHARUSAT with 9.04 CGPA. Published research at ICRAIC 2024.
Based in India
Mahesana, Gujarat. Working remotely with global teams building production-grade AI systems.
Experience
Professional journey building AI-powered systems
AI Engineer
Smart Government Technology LLC
Formerly Pulscribe.ai
- ▹Building a comprehensive hospital management platform integrating patient intake, appointment scheduling, EHR/EMR sync, and complete billing workflow.
- ▹Developing an AI-powered medical scribe that records doctor-patient encounters, transcribes via ASR, and generates structured medical documents automatically.
- ▹Designing and building FastAPI backend services with system architecture, database design, and HIPAA/PHI compliance across the entire platform.
- ▹Implementing ICD-10 code integration for insurance processing and automated billing claim submission and tracking.
- ▹Managing production workloads with Redis, Celery workers, and Prometheus monitoring in a Kubernetes-based microservices architecture.
- ▹Conducting open-source model research, testing, fine-tuning, and integration for medical domain AI capabilities.
AI/ML Intern
Smart Government Technology LLC
Formerly Pulscribe.ai
- ▹Advanced Medical Domain ASR models for clinical transcription, building end-to-end audio pipelines with noise reduction, segmentation, and domain-specific fine-tuning.
- ▹Conducted open-source model research, testing, and fine-tuning for medical domain AI capabilities, laying the groundwork for the AI-powered medical scribe.
- ▹Collaborated on early system design and architecture decisions for the hospital management platform later built out in the full-time role.
AI/ML Intern
Aiginno Pvt. Ltd.
- ▹Solely built the MVP of Ginnobot.ai — a no-code online chatbot builder platform — handling end-to-end product development including database design, backend APIs, and frontend UI.
- ▹Designed and developed custom n8n plugins and integrations, enabling seamless workflow automation directly within the platform.
- ▹Architected the full-stack Next.js application with conversational AI features, workflow orchestration, and third-party integrations for a production-ready product.
Freelance AI & Full-Stack Engineer
Upwork (Freelance)
- ▹Deliver freelance AI and full-stack engineering projects for clients on Upwork, owning the complete lifecycle from project planning and system design through development, testing, and production release.
- ▹Built and shipped WangúCRM — a WhatsApp-based CRM SaaS platform — architecting the backend, WhatsApp Business API integrations, and production deployment end-to-end.
- ▹Fine-tuned an ASR (Automatic Speech Recognition) model for the Arabic language for a client project, handling data preprocessing, audio segmentation, model fine-tuning, and evaluation — delivered with a 5-star client rating.
Projects
A collection of AI-powered applications and research projects
Pulscribe.ai – AI Hospital Management Platform
All-in-one hospital management platform for Smart Government Technology LLC. Automates intake, scheduling, EHR/EMR sync, AI medical scribing (ASR), ICD-10 coding, and billing — built HIPAA/PHI-compliant on a Kubernetes microservices stack.
WangúCRM – WhatsApp CRM SaaS Platform
WhatsApp-based CRM SaaS platform built end-to-end for an Upwork client — product planning, system architecture, backend, WhatsApp Business API integration, and production deployment.
Ginnobot.ai – No-Code AI Chatbot Builder
No-code AI chatbot builder — solo-built as MVP during my internship at Aiginno. Database design, backend APIs, frontend UI, and custom n8n plugins for workflow automation, on a full-stack Next.js conversational AI app.
F5-TTS: Gujarati Text-to-Speech Model
Fine-tuned the F5-TTS model on Gujarati speech data from IIT Madras, achieving high-quality natural speech synthesis. Trained on Kaggle GPUs for 30-40 hours with an interactive Gradio web app for real-time inference.
MedGemma 27B – 4-bit GPTQ Quantization for vLLM
Quantized Google's MedGemma 27B to 4-bit GPTQ (Marlin-optimized) for production serving on vLLM — cut VRAM from ~54GB to ~15GB and boosted concurrent throughput, at only a 3-4% accuracy drop vs. bfloat16 on clinical reasoning.
Human Detection & Counting (YOLOv8)
High-accuracy (96%) human detection model by fine-tuning YOLOv8 Nano using custom-annotated classroom images. Published as a research paper at ICRAIC'24.
SnapVid – AI Social Media Content Generator
AI-powered video generation tool using Kling API, Librosa, and MoviePy. Beat synchronization through real-time audio analysis improved visual/audio alignment by 80%. Supports 50+ concurrent requests with low latency.
VidSense – Video Recommendation Engine
Personalized video recommendation system using content-based filtering, collaborative filtering, and hybrid models. Built APIs for recommendations based on user preferences, category, and mood.
Skills
Technologies and tools I work with on a daily basis
Programming Languages
AI / ML / DL
Backend & Frameworks
DevOps & Infrastructure
Soft Skills
Achievements
Research publications, awards, and certifications
Research Publications
Enhancing Human Detection and Counting Efficiency Through Deep Learning: A Comprehensive Approach
Harsh Bhanushali, et al. (2024) • ICRAIC 2024
Proposed innovative techniques for human detection in blurred images, demonstrating the efficiency of YOLOv8-based models.
View Paper (DOI)Awards
Top 15 Teams Nationally — IndoML 2025 Datathon
Ranked among the Overall Top 15 Teams across India in Track 1 of the IndoML 2025 Datathon, organized by BITS Pilani Hyderabad Campus.
Winner — 2024 Solution Challenge
Secured 1st position (General Category) and won INR 7,500 prize for developing an innovative solution.
Certifications
- ✦Data Structures and Algorithms Using JAVA — IIT Kharagpur (NPTEL)
- ✦Introduction to Machine Learning — IIT Madras (NPTEL)
- ✦Intro to Machine Learning — Kaggle
- ✦Foundation to AI, Data Science & Data Analytics — Samatrix.io
- ✦Data Analysis Using Python — Samatrix.io
Hackathon Participations
- ✦IndoML 2025 Datathon — Organized by BITS Pilani Hyderabad Campus (2025)
- ✦InnovAItion 2025 — Organized by Intuitive.ai at DAIICT, Gandhinagar (2025)
- ✦OdooXCharusat Hackathon — Organized by Odoo India (Mar 2025)
- ✦2024 Solution Challenge Hackathon — Organized by KDPIT, CHARUSAT (Jan 2025)
- ✦IMMunoQuest Kaggle Competition — Organized by CSE, CHARUSAT (Jan 2025)
LinkedIn Activity
Latest thoughts, achievements & updates
Harsh Bhanushali
AI/ML Engineer • Dec 2025
🚀 IndoML 2025 Datathon | AI for Education 🎓 I participated in the IndoML 2025 Datathon at BITS Pilani, Hyderabad Campus. Our team Data Alchemists competed in Track 1 – AI for Education, building an AI Tutor system focused on mistake identification and pedagogically effective guidance. Proud to share we secured 15th rank in Track 1 (Mistake Classification) among strong competition from teams across India. As Team Lead, I collaborated with Manav Bagthaliya, Yash Patel & Jay Parvadiya. We also attended talks by experts from Google DeepMind, Microsoft, Netflix, Mistral AI & more — an incredible 3-day immersive experience! 🙌
Harsh Bhanushali
AI/ML Engineer • May 2025
🚀 Proud To Introduce: Our Gujarati Text-to-Speech (TTS) Model Is Live! 🎙️🇮🇳 After weeks of intense work, my teammate Harsh Ahir and I built F5-TTS — a fine-tuned Gujarati TTS model using IIT Madras Gujarati speech dataset, trained for 30–40 hours on GPU. Current performance: WER ~15%, CER ~20%. We're co-authoring a research paper and preparing public deployment on Hugging Face. Building for regional languages isn't just exciting — it's a mission to make AI inclusive for every Indian voice. Let's bring Gujarati into the AI era! 🌍
Harsh Bhanushali
AI/ML Engineer • Jun 2025
🧠💡 Turning Ideas Into Impact – SnapVid AI Wins the 2024 Solution Challenge! 🚀🏆 Thrilled to share that my teammate Rishi Patel and I secured 1st place in the 2024 Solution Challenge organized by IT-CSPIT, with a winning prize of ₹7,500! 🥇 We presented SnapVid AI — an AI-powered platform that transforms text prompts and music into beat-synced videos using Kling AI, Librosa, PyMovie, Flask & Streamlit. Huge thanks to CHARUSAT AIML Dept., Dr. Nirav Bhatt & Dr. Spoorthy V for their incredible support throughout this journey! 🚀
Harsh Bhanushali
AI/ML Engineer • May 2025
🚀 Thrilled to Share Our Latest Research Publication! Our paper "Enhancing Human Detection and Counting Efficiency through Deep Learning: A Comprehensive Approach" has been officially published in AIP Conference Proceedings! 🎉 We explore how CNNs can revolutionize human detection and counting for security, retail analytics & surveillance. Huge thanks to co-authors Harsh Maheshwari, Saumya Gohil, Gaurav Kumar & Nishant Koshti, and our HOD Dr. Nirav Bhatt for their guidance and support. 🙏
Harsh Bhanushali
AI/ML Engineer • May 2025
🎙️ Honored to Present at ICRAIC 2024! 🤖📊 I presented our research "Enhancing Human Detection and Counting Efficiency through Deep Learning" at the International Conference on Robotics, Automation, and Intelligent Computing (ICRAIC 2024), organized by RAME, Pune. Key highlights: Real-time human detection using YOLOv8, model optimization for accuracy & speed, and applications in smart surveillance systems. Stay tuned for the upcoming publication in AIP Conference Proceedings, indexed in Scopus & Web of Science! 🚀 #ICRAIC2024 #YOLOv8 #DeepLearning
Harsh Bhanushali
AI/ML Engineer • Dec 2025
🚀 IndoML 2025 Datathon | AI for Education 🎓 I participated in the IndoML 2025 Datathon at BITS Pilani, Hyderabad Campus. Our team Data Alchemists competed in Track 1 – AI for Education, building an AI Tutor system focused on mistake identification and pedagogically effective guidance. Proud to share we secured 15th rank in Track 1 (Mistake Classification) among strong competition from teams across India. As Team Lead, I collaborated with Manav Bagthaliya, Yash Patel & Jay Parvadiya. We also attended talks by experts from Google DeepMind, Microsoft, Netflix, Mistral AI & more — an incredible 3-day immersive experience! 🙌
Harsh Bhanushali
AI/ML Engineer • May 2025
🚀 Proud To Introduce: Our Gujarati Text-to-Speech (TTS) Model Is Live! 🎙️🇮🇳 After weeks of intense work, my teammate Harsh Ahir and I built F5-TTS — a fine-tuned Gujarati TTS model using IIT Madras Gujarati speech dataset, trained for 30–40 hours on GPU. Current performance: WER ~15%, CER ~20%. We're co-authoring a research paper and preparing public deployment on Hugging Face. Building for regional languages isn't just exciting — it's a mission to make AI inclusive for every Indian voice. Let's bring Gujarati into the AI era! 🌍
Harsh Bhanushali
AI/ML Engineer • Jun 2025
🧠💡 Turning Ideas Into Impact – SnapVid AI Wins the 2024 Solution Challenge! 🚀🏆 Thrilled to share that my teammate Rishi Patel and I secured 1st place in the 2024 Solution Challenge organized by IT-CSPIT, with a winning prize of ₹7,500! 🥇 We presented SnapVid AI — an AI-powered platform that transforms text prompts and music into beat-synced videos using Kling AI, Librosa, PyMovie, Flask & Streamlit. Huge thanks to CHARUSAT AIML Dept., Dr. Nirav Bhatt & Dr. Spoorthy V for their incredible support throughout this journey! 🚀
Harsh Bhanushali
AI/ML Engineer • May 2025
🚀 Thrilled to Share Our Latest Research Publication! Our paper "Enhancing Human Detection and Counting Efficiency through Deep Learning: A Comprehensive Approach" has been officially published in AIP Conference Proceedings! 🎉 We explore how CNNs can revolutionize human detection and counting for security, retail analytics & surveillance. Huge thanks to co-authors Harsh Maheshwari, Saumya Gohil, Gaurav Kumar & Nishant Koshti, and our HOD Dr. Nirav Bhatt for their guidance and support. 🙏
Harsh Bhanushali
AI/ML Engineer • May 2025
🎙️ Honored to Present at ICRAIC 2024! 🤖📊 I presented our research "Enhancing Human Detection and Counting Efficiency through Deep Learning" at the International Conference on Robotics, Automation, and Intelligent Computing (ICRAIC 2024), organized by RAME, Pune. Key highlights: Real-time human detection using YOLOv8, model optimization for accuracy & speed, and applications in smart surveillance systems. Stay tuned for the upcoming publication in AIP Conference Proceedings, indexed in Scopus & Web of Science! 🚀 #ICRAIC2024 #YOLOv8 #DeepLearning
Get In Touch
Have a project idea or want to collaborate? Let's connect!
Let's work together
I'm always excited to discuss new opportunities, AI projects, or innovative ideas. Whether you need an AI solution, backend expertise, or want to collaborate on research — reach out!