Hi, I'm Arjun Srinivas

An end-to-end system that reads a prescription and delivers the medicine: a 7-DOF Feetech arm (ROS 2 + MoveIt2, Cartesian motion with singularity-safe path rejection, Gazebo digital twin) on a mobile base with navigation and obstacle avoidance, fed by a ZED + YOLOv8 vision service over UDP, and an OCR prescription pipeline with dose scheduling via a Telegram bot and web dashboard.

An ultrasound-based directional audio system using DSP for audio modulation and a Class-D amplifier stage. Migrating legacy DSP firmware to the latest Code Composer Studio, and designing a PCB that integrates the DSP and amplifier for a smaller, higher-performance package.

An extensive literature review of multi-scale CNN architectures — analysing advances, applications and open challenges, and synthesising insights to identify research gaps and future directions in multi-scale deep learning.

A classical (non-deep-learning) face-recognition pipeline: Haar-cascade capture with Min-Max / CLAHE preprocessing, then four recognisers compared head-to-head — PCA (eigenfaces), LDA (fisherfaces), LBPH and HOG + SVM — wired into a live webcam GUI with tracking. Built the full detection, preprocessing and classifier stack.

A Streamlit app for PDF Q&A: text is chunked and embedded with Chroma (wrapped in a 256-entry LRU cache to skip redundant encoding), then cosine-similarity search retrieves the top-5 passages from Bhagavad Gita commentaries with source attribution, answered via a local Ollama LLM.

Designed a data-driven questionnaire to optimise user profiling and built the backend for LLM-based algorithmic matching, serving 150+ users based on their stated preferences.

Built and compared models (Decision Tree, Random Forest, XGBoost and more) to predict earthquake severity, reaching 98.5% accuracy through careful data cleaning, feature engineering and evaluation.

Engineered a closed-loop audio amplifier (THAT2180C VCA, LM380N PA, CA3140 VU meter) on an STM32 + VScope platform — calibrating subsystem gains/offsets, characterising PA frequency response (100 Hz–50 kHz), and writing MicroPython firmware for real-time RMS feedback and volume stabilisation.