Software engineer · ex-AMD · local-first AI
Anthony Mikinka
Co-creating a reality that is aligned with my higher selves, ambitions, and the most amount of freedom
Data science undergrad and ex-AMD co-op building local-first AI systems: multi-modal LLM apps (Lemonade Interviewer), multi-turn agent reliability, domain prompting under constraints, and multi-agent orchestration (LOOM). Strong in TypeScript/React, Python, agent architectures, and hardware-aware optimization for Ryzen AI and edge deployment.
- Background ex-AMD · ASU Data Science
- Focus Local-first agents · Ryzen AI
- Paper UCL · −29.8% tokens
Focus
- Local-first AI agents on client devices
- Multi-turn agent reliability & context management
- Hardware-aware inference (Ryzen AI / NPU / edge)
- Multi-agent orchestration & open source
- TypeScript
- React
- Python
- JavaScript
- PyTorch
- Swift
- Agent architectures
- On-device AI
Featured projects
All projects →-
Lemonade Interviewer
TypeScriptOpen-source local-first multi-modal mock interview agent (TypeScript/React): Whisper ASR, local LLM, on-device speech-out, dynamic personas, multi-phase feedback.
-
Lemonade SDK
PythonOpen-source local AI platform for running LLMs, image generation, and speech on your own GPU/NPU.
-
Universal Conditional Logic
PythonResearch + code: specification approach with 29.8% token reduction (arXiv:2601.00880).
-
LOOM
PythonMulti-agent pipeline orchestration — reliability and workflows under real constraints.
Publications
All research →Now
Full now page →-
Personal site & writing
Keeping anthonymikinka.com sharp: projects, research links, and posts when there’s something real to say.
-
Local AI & on-device ML
Continuing work on GPU/NPU tooling, inference practicalities, and open-source local AI — the same problems I cared about at AMD.
-
Research thread (UCL)
Universal Conditional Logic and related ideas — paper on arXiv, code and follow-ons as they mature.
Latest writing
All posts →-
Coho and Chinook with Dad, Frankfort
Fishing trip with Dad out of Frankfort, Michigan on August 28, 2026 — Coho and Chinook salmon on Lake Michigan. Smoked September 3.
-
Maxwell's Treatise on Electricity and Magnetism, in Python
I digitized Maxwell's 1873 Treatise into three repos: a source edition, an OCR pipeline, and maxwell-modernized — a Python library covering all 866 articles.
-
Constraint-aware systems: from data pipelines to multi-agent orchestration
Lessons from constraint-aware data pipelines (15–30% mailing cost cuts, ~70% less manual reporting) and multi-agent orchestration work including LOOM and BMAD agent templates.
-
Local-first AI agents on client devices
How I built Lemonade Interviewer—a local-first multi-modal interview agent with Whisper ASR, on-device LLMs, and multi-turn reliability under Ryzen AI constraints.
-
Universal Conditional Logic and token-efficient prompts
Universal Conditional Logic (arXiv:2601.00880): a formal specification approach that reduced token usage by 29.8% (p<0.001) via indicator functions, structural overhead modeling, and early binding.
Get in touch
Collaborations, open source, and interesting problems — email info@anthonymikinka.com, or start on Contact / GitHub.