Arjun Sasikumar
Electrical & Electronics Engineer · Applied AI/ML Builder
I build Applied AI systems, and ship production software for clients. Completed a degree across NIT Nagaland (B.Tech, EEE) and currently completing IIT Madras (BS, Data Science & Applications). Open to Forward Deployed Engineering, Solutions Engineering, and Applied AI/ML roles.
I'm an Electrical & Electronics Engineering graduate from NIT Nagaland (CGPA 9.24/10), concurrently completing a BS in Data Science & Applications at IIT Madras. My work sits at the intersection of applied AI/ML and real electrical, mechanical, and industrial systems.
I've built anomaly-detection models for ECG signals and motor bearings, and shipped industrial process-optimization tools as an Advanced Solutions Engineering intern at Yokogawa in Abu Dhabi. I have published my works in IEEE and Springer venues, and I build and ship production software as an independent consultant.
- Designed and deployed Python-based automation for oil & gas well-test accept/reject analysis, integrating rule-based logic, data-quality checks, time-series visualization, and automated reporting to reduce manual engineering review and improve decision consistency.
- Contributed to AI-driven ethane steam-cracking optimization under the Borouge AI Optimizer initiative, building conversion-prediction models and severity-aware multivariate optimization frameworks for safe Steam-to-Hydrocarbon (S/H) ratio selection.
- Developed and evaluated inferential soft-sensor models for refinery distillation (naphtha, kerosene, gas oil) on the ENOC project, applying Linear Regression, PCR, and PLS with residual diagnostics to estimate flash point from live process data.
- Building a business management platform alongside the company's website, integrating accounting and internal operational workflows.
- Working directly with stakeholders to understand requirements, design the system, and take it toward production.
- Engaged directly by a Dubai-based commercial receivables company to build and maintain their production web platform from the ground up, using Next.js, React, TypeScript, and Tailwind CSS, deployed on Vercel.
- Engineered secure backend workflows including server-side validation, input sanitization, honeypot spam protection, and rate limiting, and hardened the app with CSP, HSTS, and X-Frame-Options headers.
- Owned deployment end to end as the sole engineer, covering custom domain configuration, HTTPS/SSL, DNS, and authenticated email records, plus ongoing UI iteration and bug fixes as an independent consultant.
Problem Solving & Analytical Thinking · Systems Design & Solution Architecture · End-to-End Development · Applied AI Solution Development · Technical Communication · Customer & Stakeholder Requirement Analysis · Cross-Functional Collaboration
Vendora — Offline-First Billing & Inventory Management Platform
Kotlin · Jetpack Compose · AndroidBuilt an Android app for local shops to manage products, inventory and billing, with barcode scanning and employee accounts, designed to work even without an internet connection. Developed the system end to end, including the app, local database, authentication, cloud backup and synchronization.
Synapse: RAG-Based AI Health Chatbot
Python · FastAPI · LangChainRetrieval-Augmented Generation over a FAISS vector store, grounding an LLM (Llama-2-7B) in trusted medical sources such as the Gale Encyclopedia of Medicine, WHO, and OpenMed. Added symptom-based doctor recommendations and medication/appointment reminders. Response accuracy climbed from a 65% baseline to 91%, and retrieval latency dropped from ~500ms to under 200ms, 2.5x faster than the non-RAG baseline.
Vimra: AI-Assisted Course Generation Platform
Next.js · MongoDB · OpenAIA SaaS platform for educators and content creators to generate, manage, and monetize course content, integrating the OpenAI API for content generation and Stripe for payments.
Edge-Efficient Autoencoder Framework for ECG Arrhythmia Detection
Biomedical · ECGA lightweight unsupervised anomaly-detection framework for real-time arrhythmia detection on edge and wearable devices, requiring no labelled pathological data. Evaluated Flat VAE and LSTM-VAE architectures, then extended into a shallow Conv1D autoencoder with threshold-based anomaly detection for edge deployment.
AI-Based Bearing Fault Diagnosis Using Vibration Signals
Mechanical · VibrationBearing fault-diagnosis frameworks for electrical machines, progressing from classical ML and VAE-based approaches to 1D-CNN architectures with latent-space anomaly scoring. Includes class-conditional trust and anomaly-scoring methods for confidence quantification under uncertain operating conditions.
PPE Compliance Detection for Construction Sites (“Eagle Eyes”)
Computer Vision · SafetyA YOLO-based computer-vision system monitoring PPE compliance (helmet, vest) on construction sites in real time. Exported and benchmarked across ONNX, TensorFlow Lite, and TensorFlow.js for CPU, edge/mobile, and browser-based deployment respectively.
Multi-Modal Edge Data Processing for a Real-Time Landslide Early-Warning System
Environmental · Edge ComputingCo-authored research on processing multi-modal sensor data directly at the edge to support real-time landslide early-warning alerts.
More repositories at github.com/83Gh0st.
- “A One-Shot, Lightweight, High-Channel Shallow Edge-Ready Conv1D Autoencoder with Global Feature Summarization for Early and Efficient Diagnosis of Cardiovascular Disorders.”
- “Edge Compatible Anomaly Scoring of 1D CNNs with Confidence Quantification for Explainable Fault Classification under Uncertain Conditions.”
- “Trust-Gated Selective Classification for Bearing Fault Diagnosis via Class-Conditional Latent-Space Scoring.”
- “Edge-Optimized Multi-Class Threat Detection Using YOLO: An Occlusion-Aware, CLAHE-Enhanced Framework for Real-Time Intelligent Surveillance.”