Skip to content
View kamaladhi's full-sized avatar

Highlights

  • Pro

Block or report kamaladhi

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
kamaladhi/README.md

Hi 👋, I'm Jeevakamal K R

AI Engineer • Agentic AI • Multimodal AI • PINNs • HPC


🎓 About Me

AI & Data Science undergraduate at Amrita Vishwa Vidyapeetham (B.Tech, 2023–2027).

I build systems at the intersection of physics, machine learning, and real-world deployment — from physics-informed neural networks to production healthcare AI. Currently focused on agentic AI systems, multimodal RAG, and deploying cutting-edge models in resource-constrained environments.


💼 Recent Work

AI/ML Engineering Intern — UVS Cube Infotech, Madurai Mar 2026 – May 2026 · Completed ✅

Built and deployed two production AI systems for the MyBeat live healthcare platform:

🤖 MyBeat Care AI Chatbot

  • Hybrid RAG pipeline: FAISS (dense) + BM25 (sparse) + PubMedBERT reranking for accurate medical query resolution
  • FHIR R4 / ABDM standards integration for real-time patient history injection into LLM prompts
  • Deterministic severity scoring engine (0–100) with red-flag symptom detection
  • Triple-Gate output safety — guarantees zero diagnostic statements in all responses

Stack: FastAPI · FAISS · BM25 · PubMedBERT · Llama-3.1 · Redis · Docker · NGINX · GitHub Actions

🎙️ MyBeat Voice Assistant

  • End-to-end voice pipeline: Whisper ASRLlama-3.1 NLU (Groq) → pyttsx3 TTS with Gemini failover
  • Dual-layer NLU — keyword pattern matching (fast/offline) + Llama-3.1 for complex sentences; supports 8 intents
  • 3-Stage Cascade Recommendation Engine for doctor ranking:
    • Stage 1 (60%) — Content: Distance · Experience · Fee · Language
    • Stage 2 (25%) — Feedback: Ratings · Repeat visits · Booking conversion
    • Stage 3 (15%) — Trust: Verified credentials · Urgent availability · Insurance match
  • Voice-to-UI control — returns structured JSON uiActions (NAVIGATE, AUTO_FILL, HIGHLIGHT_RECOMMENDATION) for hands-free navigation
  • OpenWakeWord (ONNX) for always-on "Hey MyBeat" hands-free activation
  • Secured all endpoints with JWT Bearer Token (HS256) authentication

Stack: FastAPI · Whisper · Llama-3.1 · OpenWakeWord · pyttsx3 · PyJWT · Docker · NGINX · GitHub Actions CI/CD · IIS Reverse Proxy


🔬 Research Interests

  • Agentic AI Systems
  • Multimodal RAG
  • Physics-Informed Neural Networks (PINNs)
  • Healthcare AI & Medical LLMs
  • Voice AI & Speech Processing
  • Federated Learning
  • Edge AI Deployment
  • High Performance Computing (HPC)

🛠 Tech Stack

ML / AI: PyTorch · PINNs · Federated Learning · LSTMs · Attention · YOLOv8 · LoRA / QLoRA · RAG · LLMs
Voice AI: OpenWakeWord · Whisper · pyttsx3
HPC: CUDA · OpenMP · C++ · Parallel & Heterogeneous Computing
Full Stack: Python · React · FastAPI · Flask · REST APIs · MySQL · HTML5 · CSS3
Hardware / IoT: Raspberry Pi · Arduino · Blynk IoT
MLOps: MLflow · Weights & Biases · Docker · NGINX · Redis · Git · GitHub Actions
Languages: Python · C++ · JavaScript · Java (basics) · MATLAB · Scala (basics)


🚀 Featured Projects

🎥 Snake Rescue Robot

Bio-inspired robot with real-time human detection using YOLOv8. Best Paper Award at SISCON 2025.

Tech: YOLOv8, RPi, Arduino
📂 View Project

⚡ EV Range Prediction (PINN)

Physics-constrained LSTM for real-world EV range estimation. <5% RMSE, W&B tracked.

Tech: PyTorch, SUMO, MLflow
📂 GitHub

🛰️ HPC Satellite Benchmark

Serial vs OpenMP vs CUDA performance analysis on 405k satellite images.

Tech: C++, CUDA, OpenMP
📂 GitHub

🔍 FedRetinaNet

Federated object detection for power infrastructure defect detection. mAP@50: 0.91.

Tech: Python, RetinaNet, Federated Learning
📂 GitHub

🇮🇳 Tamil NLP (LLaMA)

LLaMA 3.1 8B fine-tuned for Tamil↔English translation using Unsloth.

Tech: Unsloth, LoRA, 4-bit Quantization
📂 GitHub

🏙️ Urban Microgrid Digital Twin

Predictive cyber-physical resilience framework with CNN-BiLSTM.

Tech: CNN-BiLSTM, FastAPI, Digital Twins
📂 GitHub


📊 GitHub Analytics


🏆 Achievements & Highlights

🏆 Best Paper Award — SISCON 2025
📄 3 Research Publications in IEEE & AIP journals
🚀 Production Healthcare AI Systems deployed to live platform
🎓 B.Tech AI & Data Science — Amrita Vishwa Vidyapeetham


📚 Publications


🏅 Certifications

  • Physics-Informed Neural Networks — Udemy (May 2025)
  • Google AI Essentials (Jul 2024)
  • Foundations of Cybersecurity — Google (Mar 2024)

🤝 Connect With Me

LinkedIn Email GitHub

💡 Always open to collaborations on AI systems, research, and production deployments.


Jokes Card

Pinned Loading

  1. EV-Range-prediction-using-PINNS-under-real-traffic-condition EV-Range-prediction-using-PINNS-under-real-traffic-condition Public

    This project implements a Physics-Informed Neural Network (PINN) for electric vehicle range prediction under real traffic conditions. By embedding physical constraints and traffic-aware dynamics in…

    Jupyter Notebook 2

  2. A-Predictive-Cyber-Physical-Digital-Twin-for-Urban-Microgrid-Resilience. A-Predictive-Cyber-Physical-Digital-Twin-for-Urban-Microgrid-Resilience. Public

    An intelligent digital twin framework for urban microgrids that leverages cyber-physical integration and predictive analytics to improve system resilience, reliability, and adaptive energy manageme…

    Python 1 1

  3. FedRetinaNet-Insulator-Defect-Detection FedRetinaNet-Insulator-Defect-Detection Public

    FedRetinaNet is a federated learning-based object detection framework built on RetinaNet for insulator defect detection. It enables multiple clients to collaboratively train a high-performance dete…

    Jupyter Notebook 1 1

  4. Heterogeneous-Satellite-Imaging-Benchmark Heterogeneous-Satellite-Imaging-Benchmark Public

    Experimental performance study of heterogeneous parallel computing strategies for satellite image filtering (SAT-6 dataset), featuring cache-optimized tiling, dynamic scheduling, and CUDA-based NLM…

    C++ 1 3

  5. Efficient-English-Tamil-Translation-using-Unsloth-Optimized-Transformer Efficient-English-Tamil-Translation-using-Unsloth-Optimized-Transformer Public

    English→Tamil translation model built using an Unsloth-optimized transformer with LoRA and 4-bit quantization. Developed as part of our Text Analytics semester project, including our own implementa…

    Jupyter Notebook 1 1

  6. Restaurent-Managment-System-Using-Jack-language Restaurent-Managment-System-Using-Jack-language Public

    This is a simple and a Basic project which was developed using Jack Language provided by NAND2Tetris.

    1