Computer Science (AI) student at DePaul University and AI engineer who ships rigorously engineered systems. I'm building a self-growing, brain-inspired AI engine on spiking neural networks, local LLM infrastructure, and the agent-orchestration workflow behind them, including custom Claude Code skills and hooks that keep AI agents coordinated and verified.

Focus: honest metrics, tested code, software that ships.

• Brain-inspired AI engine that grows its own neurons: when its spiking network cannot solve a problem it adds neurons, passes them through a separate "liquid" reservoir network so they learn how they relate to what it already knows, then runs a sleep cycle that prunes and saves the result to one permanent data store it can recall from.
• System design: two cooperating networks inside a single durable data lake; a safety layer that sends every robot-arm (SO-101) command through one checked path with an emergency stop; and design rules checked automatically in CI.
• Built in parallel by a team of AI agents I direct: one engine component per agent on its own git branch, peer review before merge, one integrator, and a full test battery on every merged batch. The final milestone is a whole-engine test that removes each connection to prove it is needed.
• ~390 automated self-checks plus pytest; GitHub Actions CI on Ubuntu and Windows with free-threaded Python 3.14; 350+ tagged releases.

• Claude Code skill that coordinates multiple AI agents across sessions and tools: append-only hash-chained message board, file claims, blind peer review, and an anti-gaming scorer that re-runs verification instead of trusting agent reports.
• Configures custom skills and harness hooks to improve AI workflows, e.g. a Stop hook that keeps a lead agent working instead of idling, installed into any Claude account with one idempotent command; stdlib-only with a reachable-red self-check suite.

• Builds a unified, queryable knowledge graph across multiple codebases (god nodes, community detection, inferred cross-repo bridge edges) via AST parsing and static analysis.
• Returns scoped subgraphs sized for LLM context windows instead of raw file dumps.

• Set up and run the open-source claude-code-router, self-hosted, so AI-coding-client traffic goes to a local Ollama model instead of the cloud; installed globally as a CLI and used live for cost control and privacy.

• Repeatable research → build → verify (CI) → ship loops, one safe slice per cycle: multi-agent orchestration routing subtasks by model tier, persistent cross-session memory, and knowledge-graph navigation. Nothing merges without passing the full CI gate.

• Languages: Python (Advanced), Java, JavaScript/TypeScript, C++, SQL, Assembly
• AI / Machine Learning: neural networks from scratch, deep learning, TensorFlow, PyTorch, snnTorch, spiking neural networks & neuromorphic computing, surrogate-gradient training, STDP, NLP & chatbots, GPU/CUDA training, fp16 mixed precision
• LLMs & Agents: local LLM serving (Ollama), LLM request routing, retrieval-augmented context, multi-agent orchestration, custom AI skills & hooks (Claude Code), AI build loops, prompt engineering
• Software Engineering: system design & architecture, safety-critical control paths, stdlib-only tool design, GitHub Actions CI matrices, pytest, ruff, pre-commit, test-driven development, semantic versioning & releases, packaging (PyInstaller, Inno Setup), software product testing & QA
• Web & APIs: React, TypeScript, HTML, CSS, REST APIs, OpenAI API integration
• Tooling: Git/GitHub, knowledge graphs & static analysis, VS Code, Blender scripting (sketch → 3D), Adobe Creative Cloud

Chicago, IL · reach me through GitHub