All Projects
COOP
Implements a full OOP system in C including classes, inheritance, iterators, pure_virtual macro, MISRA-based static analysis, and optimized data structures (List, Queue, Binary Tree). Includes performance benchmarks vs GObject, OOC and pure-C, image-processing add-ons, robust iterator model, ref-count mechanism, and configurable memory management.
llama.cpp
Enhancements to llama.cpp including adding missing operator kernels to SYCL/CUDA backends (ROLL, SSM_CONV, SET, CONV3D), profiling improvements, GGUF conversion changes, and enabling multimodal Eagle2-VL support. Work focuses on optimized token processing and backend-level operator coverage to improve inference throughput.
Breaking the Sorting Barrier for Directed SSSP
A breakthrough algorithm that improves upon traditional approaches for computing single-source shortest paths (SSSP) in directed graphs. Eliminates the need for sorting operations, significantly reducing computational complexity. Includes theoretical analysis, implementation optimizations, and performance benchmarks demonstrating superior scalability.
Router Verification (UVM)
A full Universal Verification Methodology (UVM) testbench for verifying a hardware packet router. Includes YAPP, HBUS, and Channel UVCs; drivers, monitors, agents, virtual sequencers; reference model; scoreboard; RAL (Register Abstraction Layer); configuration database; objection handling; multiple directed and random sequences; multi-channel coordination; coverage plan; and register tests. Environment tests routing correctness, illegal packet handling, parity errors, invalid addresses, oversized packets, multi-channel concurrency, register access, and system-level integration.
QuinnTests
An advanced testing framework designed to ensure software quality and reliability. Provides automated test execution, coverage analysis, and continuous integration support. Includes unit testing, integration testing, and performance benchmarking capabilities with detailed reporting and analytics.
AgStream
A scalable streaming data platform for processing agricultural sensor data in real-time. Enables continuous monitoring of crop conditions, weather patterns, and equipment status. Features include stream processing, time-series analysis, alerting, and integration with IoT devices for precision agriculture.
Faster maximum clique
High-performance algorithms for finding maximum cliques in graph structures. Implements advanced pruning techniques, branch-and-bound optimization, and parallel processing to handle large-scale graphs efficiently. Includes both exact and approximation algorithms with performance comparisons and scalability analysis.
TeleCruncher
A high-throughput system for processing and analyzing telemetry data from distributed systems. Handles millions of events per second with real-time aggregation, filtering, and visualization. Features include stream processing, time-windowed analytics, anomaly detection, and customizable dashboards.
AgCloud
A comprehensive cloud platform for managing agricultural operations, data, and analytics. Provides centralized storage, processing, and visualization of farm data including crop monitoring, weather integration, equipment management, and predictive analytics. Features include multi-tenant architecture, real-time dashboards, and API integrations.
Weighted Matching in a Poly-Streaming Model
Advanced graph algorithms for computing weighted matchings in poly-streaming computational models. Addresses challenges of processing large graphs with limited memory by using streaming techniques. Includes approximation algorithms, theoretical guarantees, and practical implementations for real-world graph problems.
Accelerating Graph Indexing for ANNS
A high-performance indexing system for approximate nearest neighbor search (ANNS) using graph-based data structures. Accelerates similarity search in high-dimensional spaces with improved query performance and reduced memory footprint. Features include incremental indexing, dynamic updates, and support for various distance metrics.
Parallel CPU-GPU Framework for Cost-Bounded DFS
A sophisticated parallel computing framework that leverages both CPU and GPU resources for efficient depth-first search (DFS) traversal with cost constraints. Optimizes workload distribution, minimizes data transfer overhead, and provides adaptive load balancing. Suitable for large-scale graph exploration and constraint satisfaction problems.
SpeedyCam
End-to-end camera pipeline for OV5647 sensor including configuration layer, kernel-driver integration, frame acquisition, real-time image processing (debayering, WB, gamma correction), OpenMP & NEON acceleration, logging/monitoring layer, CI pipeline with ARM cross-compilation and QEMU testing. Achieves performance significantly faster than OpenCV in multiple benchmarks.
AIExpInfra
End-to-end explainability evaluation system supporting GradCAM, IG, SmoothGrad, FusionGrad, segmentation-based explainability, metric evaluation (Sensitivity, Completeness, Non-Sensitivity, Sparsity), robustness checks, dual-channel SEM, and integration with Quantus/Captum.
Hyperconverged KV-Cache Offloading for Cost-Efficient LLM Inference
Designed and evaluated a hyperconverged KV-cache offloading layer for LLM inference, including KV-cache aware routing, full KV-cache mapping, and resource-aware policies. They benchmarked NVIDIA Dynamo vs LLM-D with and without LMCache, analyzed hit-rate vs performance trade-offs, investigated DRAM usage in the MOONCAKE runtime, and integrated LMCache with KVRocks and a custom Pliops connector. The project included Kubernetes-based deployments, GPU/CPU cache sizing calculations, and robustness improvements such as timeout handling and failure-safe clients.
SwapKV
A distributed key-value storage system with intelligent memory management and swapping mechanisms. Designed for handling large datasets that exceed available RAM by efficiently swapping data between memory and disk. Features include consistent hashing, replication, and automatic failover for high availability.
OpenVINO
Contributions to OpenVINO PyTorch Frontend: added missing operators (smooth_l1_loss, fake quant ops, quantile, linalg_multi_dot), debugging system design, INT8/FP16 benchmarking (e.g., EfficientNet-V2-XL), and integration across Model Optimizer and Runtime. Includes operator validation, code-path debugging, and cross-framework consistency improvements.
NanoVerse
A real-time infinite-world game where the world stays alive even when players disconnect. Combines GRU-based movement prediction, personalized Gemma-3 chat models with LoRA fine-tuning, DPO preference learning, chunk-based world streaming, YOLO-based perception, daily adaptation loops, and Matryoshka quantization for performance. Includes layered engine: GRU world model + personalized chat + real-time behavior engine.
onnx runtime
Fixes and enhancements across ONNX Runtime CPU Execution Provider: QuantizeLinear kernel mismatch (Issue #25362), MatMulNBits extension for 3D weight support, QDQ graph quantization bug fixes, Reduce op PRE/POST broadcasting corrections, full ONNX/NumPy-compliant broadcasting implementation for LayerNorm/RMSNorm, and multiple PRs resolving issues from 2020–2025. Includes deep tracing through kernel registration, type constraints, and execution graph selection logic.
BSDFlow
A full-scale microservices-based management platform supporting dynamic entities, groups, processes, and events. Includes multi-tenant table architecture, generic Kafka handlers with correlation-based async RPC, Redis caching for performance, real-time updates via WebSocket & SignalR, integration with LLMs, Excel watchers, advanced validation using Strategy Pattern, RBAC permissions, GIS mapping, smart search with PostgreSQL FTS+GIN+Trigrams, and a complete load-testing engine. Enables organizations to define custom workflows, attach documents, track participants, and manage system-wide behaviors across distributed services.
AgDetection
A system that standardizes model evaluation for classification, detection, and segmentation tasks. Includes ONNX-first architecture, dataset adapters, incremental evaluation engine, batch-size optimization, distributed caching, GUI comparison tools, and multi-framework support.
SemTTI
Three generative pipelines: (1) Segmentation→SEM using Pix2Pix and CycleGAN; (2) Sketch→SEM via Pix2Pix; (3) Text→SEM exploration using tagging models and textual inversion. Includes dataset building with Meta SAM-2, GAN training (CycleGAN up to 300 epochs, Pix2Pix 150 epochs), FID/KID/LPIPS/PSNR/SSIM evaluation, classifier-based realism testing, and failure-case analysis (artifacts, domain mismatch).
SemNR
A full SEM denoising workflow integrating classical filters (BM3D, Gaussian, Bilateral, NLM) with deep-learning U-Net models for Noise→Clean training. Includes custom datasets (Tin-balls real pairs, synthetic wafer images), noise simulation pipeline, multi-client FastAPI backend, PyQt6 GUI, per-stage PSNR/SSIM metrics, MinIO-based dataset consistency, pause/resume pipeline, and concurrent multi-model processing.
RAG Anything
A retrieval-augmented generation platform combining keyword search, OpenSearch indexes, embedding search, rerankers, confidence-aware pipelines, Wikipedia-scale preprocessing, LLM selection experiments, student/teacher distillation, and multi-stage QA flows. Includes Basic RAG, Query-Expanded RAG, Multi-Stage RAG + Reranker, and Multi-Stage RAG + Confidence. System optimized for precision/recall and scalable document ingestion.
Fruit
Fruit sub-project of AgCloud. A comprehensive cloud platform for managing agricultural operations, data, and analytics. Provides centralized storage, processing, and visualization of farm data including crop monitoring, weather integration, equipment management, and predictive analytics. Features include multi-tenant architecture, real-time dashboards, and API integrations.
Air
Air sub-project of AgCloud. A comprehensive cloud platform for managing agricultural operations, data, and analytics. Provides centralized storage, processing, and visualization of farm data including crop monitoring, weather integration, equipment management, and predictive analytics. Features include multi-tenant architecture, real-time dashboards, and API integrations.
Ground
Ground sub-project of AgCloud. A comprehensive cloud platform for managing agricultural operations, data, and analytics. Provides centralized storage, processing, and visualization of farm data including crop monitoring, weather integration, equipment management, and predictive analytics. Features include multi-tenant architecture, real-time dashboards, and API integrations.
Security
Security sub-project of AgCloud. A comprehensive cloud platform for managing agricultural operations, data, and analytics. Provides centralized storage, processing, and visualization of farm data including crop monitoring, weather integration, equipment management, and predictive analytics. Features include multi-tenant architecture, real-time dashboards, and API integrations.
Sound
Sound sub-project of AgCloud. A comprehensive cloud platform for managing agricultural operations, data, and analytics. Provides centralized storage, processing, and visualization of farm data including crop monitoring, weather integration, equipment management, and predictive analytics. Features include multi-tenant architecture, real-time dashboards, and API integrations.
Infrastructure
Infrastructure sub-project of BSDFlow. A full-scale microservices-based management platform supporting dynamic entities, groups, processes, and events. Includes multi-tenant table architecture, generic Kafka handlers with correlation-based async RPC, Redis caching for performance, real-time updates via WebSocket & SignalR, integration with LLMs, Excel watchers, advanced validation using Strategy Pattern, RBAC permissions, GIS mapping, smart search with PostgreSQL FTS+GIN+Trigrams, and a complete load-testing engine. Enables organizations to define custom workflows, attach documents, track participants, and manage system-wide behaviors across distributed services.
Sensors
Sensors sub-project of AgCloud. A comprehensive cloud platform for managing agricultural operations, data, and analytics. Provides centralized storage, processing, and visualization of farm data including crop monitoring, weather integration, equipment management, and predictive analytics. Features include multi-tenant architecture, real-time dashboards, and API integrations.
Workflow
Workflow sub-project of BSDFlow. A full-scale microservices-based management platform supporting dynamic entities, groups, processes, and events. Includes multi-tenant table architecture, generic Kafka handlers with correlation-based async RPC, Redis caching for performance, real-time updates via WebSocket & SignalR, integration with LLMs, Excel watchers, advanced validation using Strategy Pattern, RBAC permissions, GIS mapping, smart search with PostgreSQL FTS+GIN+Trigrams, and a complete load-testing engine. Enables organizations to define custom workflows, attach documents, track participants, and manage system-wide behaviors across distributed services.
Query
Query sub-project of BSDFlow. A full-scale microservices-based management platform supporting dynamic entities, groups, processes, and events. Includes multi-tenant table architecture, generic Kafka handlers with correlation-based async RPC, Redis caching for performance, real-time updates via WebSocket & SignalR, integration with LLMs, Excel watchers, advanced validation using Strategy Pattern, RBAC permissions, GIS mapping, smart search with PostgreSQL FTS+GIN+Trigrams, and a complete load-testing engine. Enables organizations to define custom workflows, attach documents, track participants, and manage system-wide behaviors across distributed services.
Events
Events sub-project of BSDFlow. A full-scale microservices-based management platform supporting dynamic entities, groups, processes, and events. Includes multi-tenant table architecture, generic Kafka handlers with correlation-based async RPC, Redis caching for performance, real-time updates via WebSocket & SignalR, integration with LLMs, Excel watchers, advanced validation using Strategy Pattern, RBAC permissions, GIS mapping, smart search with PostgreSQL FTS+GIN+Trigrams, and a complete load-testing engine. Enables organizations to define custom workflows, attach documents, track participants, and manage system-wide behaviors across distributed services.
Documents
Documents sub-project of BSDFlow. A full-scale microservices-based management platform supporting dynamic entities, groups, processes, and events. Includes multi-tenant table architecture, generic Kafka handlers with correlation-based async RPC, Redis caching for performance, real-time updates via WebSocket & SignalR, integration with LLMs, Excel watchers, advanced validation using Strategy Pattern, RBAC permissions, GIS mapping, smart search with PostgreSQL FTS+GIN+Trigrams, and a complete load-testing engine. Enables organizations to define custom workflows, attach documents, track participants, and manage system-wide behaviors across distributed services.
Entities
Entities sub-project of BSDFlow. A full-scale microservices-based management platform supporting dynamic entities, groups, processes, and events. Includes multi-tenant table architecture, generic Kafka handlers with correlation-based async RPC, Redis caching for performance, real-time updates via WebSocket & SignalR, integration with LLMs, Excel watchers, advanced validation using Strategy Pattern, RBAC permissions, GIS mapping, smart search with PostgreSQL FTS+GIN+Trigrams, and a complete load-testing engine. Enables organizations to define custom workflows, attach documents, track participants, and manage system-wide behaviors across distributed services.
NextOptAI
An advanced AI system for optimizing computational workflows and resource allocation. Uses machine learning to predict optimal configurations, reduce latency, and improve efficiency. Features include predictive analytics, automated tuning, and adaptive learning for continuous improvement.