NS
Nayant Srivastava
01Initiatives // Long-term

Engineering Initiatives & Systems

Active frameworks, custom engines, developer communities, and long-term world-building architectures.

AI / ML · Systems·AI / ML
Active

MINDLITE

Efficient AI / ML experimentation & lightweight edge runtime.

An engineered framework built from the ground up for high-throughput machine learning experimentation, systematic ablation sweeps, and quantized deployment on constrained edge hardware.

Design Architecture & Motivation:
  • ▸Addresses the developer friction of bloated ML frameworks by focusing on reproducible benchmark runs and minimal runtime dependencies.
  • ▸Includes optimized C++ tensor kernels and SIMD accelerated post-training INT8 quantization pipelines.
  • ▸Designed to power real-time perception models and autonomous control systems like ADCTM.
Technical Pillars:
01.Modular experiment harness with automated hyperparameter sensitivity visualization
02.Zero-overhead C++ inference runner with contiguous memory layout
03.Hardware profiler measuring memory bandwidth saturation and cache hit rates
04.Automated baseline comparison against standard PyTorch and ONNX Runtime exports
Inference Footprint< 15MB
Latency Speedup3.4x vs PyTorch
Telemetry Throughput45k samples/s
PythonC++20PyTorchINT8 QuantizationSIMDOpenMP
Source
Graphics · Systems·Graphics
Development

OMNIX ENGINE

Experimental 3D simulation and rendering engine in C++ & Vulkan.

A custom, ground-up high-performance graphics engine written in modern C++20 utilizing the Vulkan graphics API, targeted at real-time computational geometry, physics simulation, and high-fidelity rendering.

Design Architecture & Motivation:
  • ▸Explores modern explicit GPU APIs, low-overhead memory allocators (VMA), and compute-driven rendering pipelines.
  • ▸Implements a modern render graph architecture managing explicit resource transitions and pass dependency scheduling.
  • ▸Acts as the foundational technology testbed for future simulation projects and interactive virtual worlds.
Technical Pillars:
01.Vulkan 1.3 pipeline with dynamic rendering, bindless textures, and push descriptors
02.Custom linear and pool memory sub-allocators with staging ring buffers
03.Compute shader particle simulation pipeline handling 500k+ dynamic agents at 60+ FPS
04.Data-oriented Entity Component System (ECS) architecture ensuring high cache locality
Frame Time< 2.1ms
Particle Capacity500,000+
API OverheadNear Zero (Indirect)
C++20Vulkan APIGLSL / SPIR-VGLMVMACMake
Source
Education · Engineering·Education
Building

BYTELOGIC

Technical learning & computational knowledge platform.

A first-principles technical education platform for understanding complex concepts across Computer Science, Artificial Intelligence, Computational Mathematics, and Distributed Systems.

Design Architecture & Motivation:
  • ▸Replaces static textbooks and opaque video tutorials with interactive computational sandboxes and rigorous mathematical proofs.
  • ▸Features custom 3D ASCII manifold visualizers, real-time Voronoi canvas renderers, and structured curricula across 8 foundational domains.
  • ▸Built around the pedagogy: Deconstruct -> Formalize -> Visualize -> Implement -> Stress Test.
Technical Pillars:
01.3D ASCII projection engine rendering mathematical manifolds directly in the DOM buffer
02.Interactive 2D canvas sandboxes (e.g., K-Means Voronoi partitioning, state machines)
03.LaTeX / KaTeX mathematical rendering coupled with tabbed from-scratch code implementations
04.Integrated ⌘K global concept search with live dependency graphs across prerequisites
Curricula Domains8 Domains
Learning Paths7 Structured Paths
Interactive DemosLive on /bytelogic
Next.js 15TypeScriptHTML5 CanvasKaTeXTailwind CSSFramer Motion
SourceLaunch
Community · Open Source·Community
Active

ATTACK ON CODE

Engineering community, collaborative builders & open-source culture.

A technical community founded to unite ambitious engineering students, systems developers, and researchers. Focused on disciplined software development, competitive hackathons, and open-source contribution.

Design Architecture & Motivation:
  • ▸Empowers young engineers to move beyond surface-level web tutorials toward low-level systems, mathematical thinking, and rigorous engineering practices.
  • ▸Organizes collaborative coding sprints, architecture review sessions, and open-source project development.
  • ▸Serves as a launchpad for university talent entering research labs and cutting-edge software engineering roles.
Technical Pillars:
01.Community sprint framework focusing on production-grade code reviews and git workflows
02.Systems engineering tracks covering C++, Rust, and Linux systems internals
03.Mentorship pipelines bridging academic CS curriculum with real industry architectures
04.Collaborative open-source repositories with benchmark verification suites
Engineers Mentored100+
Sprints CompletedQuarterly
Community StatusGrowing Active
Community LeadershipOpen SourceGit WorkflowsSystems Mentorship
Source
AI / ML · Architecture·AI / ML
Development

NEVULA

Autonomous agent framework & deterministic tool-use runtime.

A modular, high-reliability framework for autonomous AI agents capable of planning, self-verification, and executing deterministic tool pipelines without catastrophic hallucinations.

Design Architecture & Motivation:
  • ▸Designed to handle complex multi-step developer workflows, automated code refactoring, and benchmark verification loops.
  • ▸Implements structured state transitions, sandbox isolation, and reactive event-driven messaging between parent and subagents.
Technical Pillars:
01.Graph-based state machine enforcing pre-conditions and post-conditions on tool actions
02.Sandboxed process executor with strict timeout guards and stdout telemetry parsing
03.Hierarchical multi-agent delegation with structured conversation checkpointing
Tool Success Rate> 94%
State Determinism100% Audited
TypeScriptPythonNode.jsLLM Tool CallingState Machines
Source
Game Development·Game Technology
Long-term

LAST TRANSISTOR

Long-term AAA game development initiative & engineering journal.

A multi-year ambitious creative and engineering initiative building an expansive, immersive sci-fi game world powered by custom simulation engines and low-level computational architecture.

Design Architecture & Motivation:
  • ▸A long-term vector combining world-building, narrative design, custom graphics pipelines, physics engines, and procedural generation.
  • ▸Documented openly through an ongoing developer journal detailing every architectural decision, engine failure, and algorithmic breakthrough.
  • ▸Represents the ultimate synthesis of all engineering disciplines: systems, graphics, mathematics, AI, and art.
Technical Pillars:
01.Continuous development journal recording systems failures, math derivations, and engine rewrites
02.Custom voxel and procedural terrain generation with continuous Level of Detail (LOD)
03.Hierarchical AI behavior trees controlling autonomous entities in simulated environments
Timeline HorizonMulti-Year
Development PhaseArchitecture & Engine
C++Vulkan / Custom EngineProcedural GenerationPhysics Integrators
Source

Last Transistor: Architecture Journal

Long-term Horizon

An open engineering log detailing custom engine milestones, mathematical breakthroughs, and failure autopsies.

Living Document
PHASE 01 // Q1 2026 — Present

Omnix Vulkan Engine Integration & Compute Physics

In Progress

Coupling the game world directly with Omnix Engine: bindless GPU resource indexing, dynamic push descriptors, and compute-driven particle fields for environmental physics.

Algorithmic Breakthrough:

Implemented GPU frustum culling via indirect multidraw commands, reducing CPU draw call dispatch overhead to under 0.8ms per frame across 10,000 instanced props.

Core Technical Challenge:

Handling GPU memory fragmentation when streaming high-resolution textures into memory pools during rapid player movement.

Vulkan 1.3GLSL ComputeVMATracy Profiler
Recorded in internal dev logs