Recent project
Scheduled orchestration in n8n pulls ArXiv and Reddit through dedicated agents, then classifies and enriches items via local HTTP services in Docker. Outputs land as structured AI/ML research rows with summaries, topic tags, and relevance scores for filtering and downstream workflows.
Data science projects
Recreated and extended Gapminder-style views of population, fertility, child mortality, and poverty using country-level tables, joins, time-series analysis, and comparative visualizations.
Analyzed cardiovascular outcomes across observational and randomized studies, using simulation-based hypothesis testing, p-values, relative risk, bootstrap intervals, and hazard-rate comparisons.
Built and evaluated a k-nearest-neighbors classifier for comedy versus thriller screenplays using bag-of-words features, correlation analysis, train/test splits, distance metrics, and feature selection.
GitHub feed
Additional public repositories from mmiles10.
Fetching public repositories from `mmiles10`.
About
Physics graduate from UC Berkeley with a background in the intersection of semiconductors, data, and macro finance. Prior internship in financial planning and analysis at fintech Acorns, including AI automation for financial strategy, market research, and SQL frameworks. Builder and operator for early teams, focused on turning complex systems into useful tools across energy infrastructure, financial markets, product, operations, and AI-enabled execution.
Thinking
A short shelf on AI systems, agents, technical execution, and market direction.
IBM
A clean overview of the major machine learning categories and where each one is useful, helpful for staying grounded when AI conversations get fuzzy fast.
IBM
A useful explanation of how retrieval, augmentation, and generation fit together, plus a clear distinction between RAG and fine-tuning.
LangChain
Practical guidance on isolating agent execution, including per-conversation sandbox lifecycles and the tradeoffs of running the agent inside the sandbox itself.
GitHub
Anders Hejlsberg on why TypeScript became the default for large codebases: confidence, maintainability, and better scaling for serious product work.
Vercel
A plain-language tour of Vercel as a frontend cloud: shipping the user-facing layer from Git with minimal ops, deploy previews on pull requests, how Next.js ties in, and how the platform extends into serverless functions, analytics, and managed storage and databases.
Anthropic
A strong architecture-oriented piece on where agent patterns are actually useful, and how to think about workflows, tooling, and implementation boundaries.
NEMO Committee
The official public description of the algorithm used in European day-ahead market coupling, useful for understanding how cross-border capacity, bids, constraints, and welfare optimization translate into clearing prices.
Anthropic
One of the better explanations of why agent evals are harder than single-turn evals, and why tasks, graders, trials, and transcript review matter in production.
LangSmith
A practical look at tracing, debugging, evaluating, and monitoring agent behavior once you move beyond local demos.
Khosla Ventures
A long-form investor view on the range of AI outcomes, useful less for certainty than for pressure-testing first-principles assumptions about where this all leads.
Contact
Reach out about roles, projects, research, product work, or early-stage teams that need someone who can make useful things happen.
