Projects

Neural Networks in Complex Spaces

Tags: Complex-Valued Neural Networks, RF Modulation, Representation Learning, PyTorch

Research code for complex-valued neural networks in PyTorch, with layers, activations, baselines, and benchmark experiments across synthetic RF, physics, and neuro-style tasks. Companion to the paper on when CVNNs help under representation and optimization constraints.

Code | Paper

AI Chip Design Roadmap

Tags: AI Accelerators, GPU Programming, Hardware Design, Curated Resources

A curated roadmap from digital logic and GPU programming through AI accelerator design, ML compilers, and production chip architectures, with papers, tools, and open-source tapeout resources.

Code

Edge Vision-Language Model Server

Tags: Edge Inference, Vision-Language Models, YOLOv8, Open Source

Free, open-source edge inference server combining YOLOv8 object detection with vision-language reasoning for deployable multimodal perception pipelines.

Code

Decision Support System for Global Vaccine Funding

Tags: Public Health, Optimization, Agent-Based Simulation, Master's Thesis

Master's thesis project at RIT: a decision-support system for global vaccine funding allocation using data-driven and simulation-based approaches.

Code

3D Gaussian Splatting for Novel View Synthesis

Tags: Neural Rendering, 3D Reconstruction, Novel View Synthesis, Computer Graphics

A PyTorch implementation of 3D Gaussian Splatting for real-time radiance field rendering and novel view synthesis. Features end-to-end training pipeline, inference capabilities for rendering novel camera trajectories, and support for Mip-NeRF 360 datasets. Implements learnable 3D Gaussian primitives with spherical harmonics for view-dependent appearance modeling.

Code

Vision-Language Models

Tags: Multimodal AI, Vision-Language Understanding, Deep Learning

Research and implementation of vision-language models exploring the intersection of computer vision and natural language processing. Focuses on developing and evaluating models that can understand and reason about both visual and textual information for downstream tasks.

Code

Self-Driving Negotiator

Tags: Autonomous Vehicles, Game Theory, Multi-Agent Systems, Decision Making

An interactive simulation exploring game theory principles in autonomous vehicle decision-making and negotiation strategies. Demonstrates how game-theoretic approaches can be applied to multi-agent scenarios in self-driving systems, enabling vehicles to make strategic decisions in complex traffic situations.

Demo

Uncomputable Oracle Optimizer

Tags: Bayesian Optimization, Computability, Risk-Aware AI

A novel Bayesian optimization framework for black-box functions with uncertain or uncomputable outputs, integrating Gaussian process surrogates and halting probability modeling for risk-aware decision making in computationally intractable domains.

Code

Perceptual Piercing

Tags: Vision Enhancement, Selective Dehazing, Object Detection

An adaptive deep learning pipeline for object detection in adverse weather, leveraging selective region dehazing and human visual system-inspired attention to maintain accuracy in fog, haze, and clear conditions.

Code | Paper

MoonMetaSync

Tags: Lunar Image Registration, SIFT+ORB Fusion, Space Imaging

A hybrid feature-based image registration toolkit for lunar surface analysis, combining SIFT and ORB descriptors to achieve robust alignment across diverse lunar imagery and mission datasets.

Code | Paper

CONVERT: COvid Narrative Evolution and Remorse Tracking

Tags: NLP, Media Analysis, Vaccine Remorse, COVID-19

A natural language processing framework for tracking COVID-19 vaccine remorse and narrative evolution in media, using sentiment analysis and temporal topic modeling to inform public health interventions.

Code

Adaptive and Robust Sensor Fusion in Autonomous Vehicles

Tags: Autonomous Vehicles, Multi-Modal Fusion, Robust Perception

Evaluation and comparison of advanced sensor fusion architectures (Spatial-IL, Fusion-DETR, HFN) for robust perception in autonomous vehicles, with adversarial testing on the KITTI dataset.

Code

Hierarchical Prompting Taxonomy (HPT)

Tags: LLM Evaluation, Cognitive Taxonomy, Prompt Engineering

A universal evaluation framework for large language models, introducing the Hierarchical Prompting Taxonomy (HPT) and Index (HPI) to benchmark cognitive complexity and prompt-based reasoning.

Code | Paper

IndicMMLU-Pro

Tags: Multilingual Benchmark, Indic LLMs, Language Understanding

A comprehensive benchmark for evaluating large language models on multi-task language understanding across major Indic languages, supporting research in multilingual NLP and model generalization.

Dataset | Paper

More open-source work on GitHub:

github.com/ashu1069