AI systems, backend platforms, security engineering, local-first applications, client delivery, and demoscene productions. Each entry focuses on architecture, evidence, and responsibility - not only a feature list.
Recent work
Two shipped demoscene projects: an award-winning learning tool and a real-time audiovisual production.
Fix It Competition · Assembly Summer 2026
2nd place · 132 points
A demoscene builder for beginners. Export your first real-time browser demo in 5 minutes.
Built in 20 hours with 21 learning paths, real-time effects, offline export, and no backend. Won 2nd place at the Assembly Summer 2026 Fix It Competition among 13 finalists.
GitHub pending · not yet open source
AI Coding (Vibe Demo) Competition
Assembly Summer 2026
Real-time audiovisual demo (122.5s @ 1080p60) exploring AI through procedural graphics.
A deterministic 60fps engine with FFT visualization, synchronized narration, procedural effects, and full AI tool disclosure.
AI engineering
Agent infrastructure, evidence-gated design workflows, RAG pipelines, and AI-assisted engineering tools.
Open source · v2.0.3 published
A local-first Agent Skill, MCP server, CLI, and deterministic UI quality gate.
An independently authored design-engineering system that converts product intent, truthful interface state, information hierarchy, architecture, accessibility, and browser behavior into enforceable contracts. It provides 20 scoped MCP tools, BM25 knowledge retrieval, repository audits, Chromium verification, evidence-separated reviews, bounded repair, offline packaging, and clean-room independence checks. The v2.0.3 release passed 168 automated tests and qualified contained delivery pilots across SceneStart, AegisOPS, and Azure Optimizer without presenting automated evidence as representative-user validation.
Public demo
4-agent LangGraph pipeline (AMD Hackathon 2026): MITRE ATT&CK to Sigma rules and SOC playbooks. Hugging Face Spaces deploy with demo mode default; live vLLM/ROCm path for full inference.
AI code-quality and security auditor with FastAPI endpoints, RAG over 20,875 embeddings, PostgreSQL/pgvector, Redis caching, LangGraph tool use, PII/secrets redaction, structured outputs, audit logging, retention controls, and repeatable evaluation.
AI-assisted system for analyzing cloud infrastructure, detecting inefficiencies, and generating actionable cost and configuration optimizations.
A fully local speech pipeline: faster-whisper speech recognition, Ollama local inference, Piper text-to-speech, and a FastAPI boundary. Designed as an end-to-end systems integration project without cloud speech dependencies.
Local AI engineering
A local coding-agent track around a reconstructed CLI, Qwen2.5-Coder fine-tuning experiments, LoRA adapters, merged model artifacts, GGUF packaging, and an Ollama-compatible local inference path.
Local-first tool
A browser roadmap and routing application with deterministic offline fallback and an optional Cursor-compatible chat proxy. Built to make AI engineering work visible as a navigable system rather than a list of prompts.
Active work · Vanguard public / Fathom private
A scope-bound bug-bounty/red-team workflow with evidence capture, report drafting, local RAG foundations, and a separate defense guard for process health, kill controls, heartbeats, and operator alerts.
Fathom remains private; Vanguard is the public companion guard and operations component.
Machine learning
Work that shows how the systems are reasoned about, evaluated, and built underneath the application layer.
Coursework · grade 5/5
Comparative convolutional-network experiments across local CUDA, CPU TensorFlow/Keras, Google Colab Tesla T4, and AMD ROCm MI300X environments, varying architecture, activations, optimizers, batch sizes, and class subsets.
Coursework · grade 5/5
Python implementations and experiments from algorithm-design coursework, translating formal problem-solving techniques into tested, runnable solutions.
Coursework · in progress
Current university coursework implementing advanced algorithm-design problems in Python, including dynamic programming, string algorithms, graph-style problems, and combinatorial techniques.
Repository not published yet.
Applications
End-user software and production systems with real data models, APIs, deployment, and operational boundaries.
Private client project
Private client delivery: work orders linked to inventory with stock deduction, remnants, and audit trail. FastAPI, React, PostgreSQL, nginx, systemd.
Client identity and production details withheld for confidentiality. Sanitized portfolio demo available on GitHub.
Two implementations of the same real-time league problem: a React/Node/Prisma version with SSE, and a Python-first FastAPI/SQLAlchemy/SQLite version with async synchronization, validation, REST/SSE boundaries, and a server-rendered dashboard.
A local-first product suite spanning Career, Ledger, Debts, Health, PDF conversion, and video downloading. Shared design tokens, browser storage, Python services, local HTTPS, systemd units, health checks, metrics, authentication, and production-oriented runtime safeguards.
Live production · iterative rebuild
A portfolio platform that records the move from handcrafted static pages to modern server-rendered application architecture.
The first version was a multi-page HTML/CSS/JavaScript site with reusable header and footer components, security-focused project sections, and manually managed assets. The next rehaul introduced utility-first Tailwind, responsive navigation, richer metadata, and deployment hardening. A Gemini/AI Studio branch explored Angular 21 SSR, Angular Material/CDK, Tailwind, Motion, and a Gemini API integration path. The current version uses TanStack Start, React, TypeScript, Tailwind CSS, Nitro/Node.js SSR, route-level SEO metadata, responsive navigation, and local asset bundling. The delivery workflow runs through Git: changes are committed and pushed, fetched into the deployment workspace, reviewed and approved, then built and applied to production through cPanel Passenger.
Source repository private; the live platform and its architecture are public.
Client work · in progress
Current website projects combining visual design, content systems, integrations, and maintainable delivery workflows. Client details remain limited while the work is in progress.
Client work · in progress
A Finnish community platform with a themed visual system, streamer and team pages, live Twitch status, Discord application workflow, private moderation review, privacy/terms pages, and server-side integrations.
Client project; public details and deployment URL are intentionally limited during development.
Client work · in progress
A WordPress website restoration and design workflow using Docker Compose, PHP 8.2, MariaDB, migration/archive handling, theme and custom CSS work, responsive refinement, and repeatable local restoration before design changes are applied.
Production archive and user data remain private while the redesign is in progress.
Security engineering
Defensive engineering, controlled security labs, browser tooling, and scope-aware research workflows.
Private lab infrastructure
A mission-control interface for authorized research: program scope, prompts, gated workflows, HexStrike chains, Caido integration, job state, and hunt output. The portfolio story is the orchestration and safety boundary, not the upstream tooling.
Private infrastructure; no live targets, credentials, or operational data are published.
Sanitized public repository
Generalized proof-of-concepts and technical writeups covering vulnerability classes such as injection, validation, CSRF, CORS, file handling, GraphQL edge cases, and decompression handling, with targets and identifiers removed.
Elastic-based telemetry and detection workflows spanning Windows and Linux endpoints: collection, normalization, log shipping, rules, investigation paths, and realistic noise-aware validation.
Controlled protocol-level experiments that pair attack mechanics with detection concepts, visibility, and mitigation thinking. The emphasis is understanding network behavior in authorized lab environments.
Small Python utilities for defensive analysis, endpoint visibility, security-state inspection, and early-stage blue-team workflows, designed to be understandable and extensible rather than opaque scanners.
Multi-language exercises covering memory safety, cryptography, authentication, file handling, validation, permissions, process behavior, undefined behavior, and system hardening—with unsafe patterns paired to safer implementations.
Professional recommendation
A testimonial from The Library of Alexandria project, where I worked as a technical consultant and TrueNAS systems builder. The work covered the architecture, assembly, configuration, and operational handover of the media-archive storage system.
The Library of Alexandria · TrueNAS architecture · March–July 2026
More work
Additional public utilities, browser tools, labs, and research repositories. Together, these show the breadth behind the featured systems: AI engineering, software delivery, security analysis, and operational tooling.