Computer Science student exploring how intelligent systems reason, plan, and verify — not just generate more tokens.
I'm a Computer Science student interested in AI reasoning, research engineering, and reliable agentic systems. I build research tools, interactive experiments, and software workflows that make complex ideas easier to inspect, test, and reproduce.
My focus is on evidence over appearance, verification over assumption, and useful systems over impressive demos.
A map of my public work across AI research, agent engineering, specialist toolkits, and applied systems. Each project name opens its repository.
- ARGUMENTUM Research Engine — A self-hosted, multi-agent research pipeline for literature discovery, evidence mapping, methodological critique, and auditable research outputs.
- ReasoningAI Studio — An evidence-first research toolkit with portable skills, reusable prompts, templates, and local utilities for reproducible research.
- Neural Networks Playground — An interactive machine-learning lab for training small neural networks and inspecting activations, gradients, optimizers, and decision boundaries.
- Algorithm DSA Forge — A visual learning environment for data structures, algorithms, pathfinding, machine learning, and AI reasoning, with step-by-step execution.
- The Startup — Gemini CLI Edition — A Gemini CLI-native engineering workflow with specialist agents, spec-driven development, adversarial reviews, and security and cost gates.
- LaunchCrew — A Codex-first, spec-driven toolkit for turning product ideas into scoped plans, delegated implementation, verified changes, and clear handoffs.
- Claude Garage — A Claude Code plugin that brings structured startup-style workflows to planning, parallel building, testing, auditing, and shipping.
- Agent Workbench — A portable library of agent skills, role blueprints, and templates for software engineering, research, and delivery.
- AgentFieldkit — Evidence-first playbooks for repository exploration, software engineering, security, AI/ML research, and multi-agent collaboration.
- Implicit State Machines — Agent workflows for finding hidden application lifecycles and refactoring fragile, boolean-heavy state into explicit, testable models.
- Agentic Skill Garage — Creative-production workflows for AI agents, supported by local Python tools for asset preparation, media operations, and deliverable checks.
- JaneStreet Agentic Skills — Specialist playbooks for quantitative trading, market microstructure, order-book systems, low-latency engineering, risk controls, and research validation.
- KINETIQ Machine Simulation Skills — Portable skills for interactive 3D machine explainers, mechanical simulation, physics validation, and offline delivery.
- Pixelcraft — A design-quality skill library that helps agents detect generic interfaces and improve hierarchy, typography, layout, accessibility, and consistency.
- MarketPilot AI — A model-agnostic trading research toolkit for market scans, risk checks, scenario replay, walk-forward evaluation, and paper-only workflows.
- The Sharks — A multi-agent startup evaluation panel that pressure-tests customers, revenue, unit economics, competition, and founder-market fit.
- Kinoquence — A local-first, model-agnostic production workflow for animated shorts, with structured storyboards, asset tracking, captions, credits, and quality checks.
- SheerFlow — SheerID Workflow Studio — A JavaScript workbench for authorized SheerID API integrations, verification workflows, document uploads, and auditable job tracking.
- CampusFlow — A classroom-attendance prototype with rotating QR check-ins, duplicate prevention, and CSV reporting; built around fictional demonstration data.
- HoneyPot Guard — A local-first pre-commit scanner for leaked secrets, suspicious outbound endpoints, and risky code patterns.
- Claude Bundle Domain Inspector — A small macOS utility for inspecting narrowly scoped domain strings in Claude Code bundles, without executing the inspected binary or making network requests.
- MAPIE — A repository copy of the upstream MAPIE project, a scikit-learn-compatible library for conformal prediction and risk control. Listed as an ecosystem reference, not as original authorship.
Languages and Runtime
AI, Machine Learning, and Research
Agentic Development
Engineering and Workflow


