Course: System-on-Chip (SoC) Platforms (Prof. Luca Carloni, Columbia University)

Timeline: Oct 2025 - Nov 2025

This project involved architecting and implementing a synthesizable hardware accelerator for the BERT-Tiny language model in SystemC. The design covered all 11 layers of the inference pipeline, including multi-head attention, feed-forward networks, and GELU activation.

Course: Heterogeneous Computing for Signal and Data Processing (Prof. Zoran Kostic, Columbia University)

Timeline: Oct 2025 - Nov 2025

Developed a high-throughput, GPU-accelerated LDPC decoder using PyCUDA (or PyOpenCL). The goal was to leverage parallelism for real-time signal processing in 6G wireless communication systems, with a specific focus on optimizing for low Signal-to-Noise Ratio (SNR) scenarios.

Timeline: Jan 2025 - Mar 2025

Implemented dynamic resource allocation algorithms with PyTorch, achieving 92.3% accuracy while reducing the memory footprint by 83.4% and latency by 74.5% compared to the original teacher models.

Engineered modality-specific compression techniques and lightweight feature alignment strategies. This enabled the project to be deployed with adaptive resource management, maintaining 82.8% accuracy with a 9.3ms inference latency.

Collaboration: Joint Project with the Infectious Diseases Dept, Christian Medical College, Vellore & VIT

Timeline: Sep 2024 - Dec 2024

Built an Intelligent Positioning System that leverages the sophisticated Bowyer-Watson triangulation algorithm for accurate location mapping, using Bluetooth Low Energy (BLE) and Wearable-Health technologies.

Leveraged Machine Learning techniques, such as "Ensemble-based Boosted Decision Trees," to analyze and predict the likelihood of HAIs among medical professionals as a pre-emptive prevention measure.

Timeline: Aug 2024 - Dec 2024

Developed a novel hybrid deep learning model for EEG-based emotion recognition by integrating Stacked Autoencoders, LSTM networks, and temporal sequence learning. The model was built on the DEAP dataset from Queen Mary University, London.

Utilized MNE and SciPy for EEG preprocessing and feature extraction, resulting in 89% accuracy in Valence-Arousal prediction under 10-fold cross-validation, outperforming baseline SVM and LSTM models.

Timeline: Apr 2023

• Designed and Implemented an Image Processing software using Machine Learning and 8051 Microcontrollers.
Link to codes can be found here.

Timeline: Oct 2022 - Feb 2023

• Created a novel Attitude Flight control system using moments produced by Gyro-spheres.
• Worked as Team Leader of the CS Department, in charge of developing the Flight Control GUI.

Timeline: Aug 2023 - Dec 2023

• Deployed an IoT integration utilizing LoRa and ML principles to leverage LLMs in IoT hardware communications via API calls to promote multilingual and extensive irrigation control among local farmers, boosting agricultural eco-efficiency.
• Volunteered to serve as a liaison between the "Artificial Intelligence Division" and the "Agriculture Technology Division" handling team management toward seamless integration and collaboration.
• Conducted 2 rounds of field tests to measure precise soil moisture data interpretation capabilities and personalized irrigation recommendations, tailored to specific crop and soil requirements via the AgWise platform.

Organization: iNeuron.ai

Timeline: Jun 2023 - Jul 2023

• Designed and developed a Predictive Maintenance solution for implementation in the Aerospace industry using Machine Learning Technologies specifically working on API Interfacing and Datasets.