Case Studies

How I make software trustworthy enough to ship, and how I get real work delivered with coding agents. Each one covers the problem, the constraints, what I built, how it's verified, and what happened.

Evaluation & test infrastructure

Test suites that prove a safety rule rather than sample it, offline fidelity harnesses checked against independent ground truth, and evidence pinned to the exact commit under review.

Delivery with coding agents

Working software shipped through agent-written pull requests, kept honest by review contracts, automated review gates, and written rules that every agent session reads.

Evaluation & test infrastructure Delivery with coding agents August to October 2026

AstroAI: Social Publishing That Never Double-Posts

At-most-once delivery, immutable approvals, and safety tests that can't be broken by the calendar.

An unattended social publisher for AstroAI that treats a public post as irreversible. Every post needs a sealed, append-only human approval of its exact bytes, and any send whose outcome can't be proven is held for a person instead of retried.

tests across the three publish-safety suites: 128; tests across the three publish-safety suites
automatic retries once a send's outcome is ambiguous: 0; automatic retries once a send's outcome is ambiguous
tests a calendar date once broke; now clock-independent: 23; tests a calendar date once broke; now clock-independent
clock shifts both safety suites were run under: -10 to +100 yrs; clock shifts both safety suites were run under

Read the case study Private repo Live site

Evaluation & test infrastructure Delivery with coding agents July to October 2026

AstroAI: Review Evidence for Agent-Built Pull Requests

Proof of what shipped: exact-head evidence reports, seeded private previews, and a PR contract checked in CI.

Review infrastructure for a product built mostly through coding-agent PRs. An offline Playwright harness produces self-contained visual evidence pinned to the exact head commit, and a PR contract enforced in CI requires a private live preview plus a manual review plan for that specific feature.

PRs merged Jul 1 to Oct 31, 2026: 171; PRs merged Jul 1 to Oct 31, 2026
unit and integration tests passing on main: 2,814; unit and integration tests passing on main
offline browser end-to-end tests passing on main: 62; offline browser end-to-end tests passing on main
commits on main with a Claude co-author trailer: 125 / 306; commits on main with a Claude co-author trailer

Read the case study Private repo Live site

Evaluation & test infrastructure July 2026

AstroAI: An Evidence-Backed Chart Engine Cutover

A migration where the evidence came first: 15 edge-case charts, an independent reference, and a one-line rollback.

Moved AstroAI's natal chart computation from a paid hosted API to local Swiss Ephemeris math. The cutover was approved on an offline fidelity harness that checks every planet, angle, and house cusp of 15 edge-case charts against NASA JPL Horizons and independent ephemeris calls.

worst planet error vs NASA JPL Horizons (bar: 60"): 1.9"; worst planet error vs NASA JPL Horizons (bar: 60")
edge-case charts compared element by element: 15; edge-case charts compared element by element
house cusp difference vs direct Swiss Ephemeris: 0.000"; house cusp difference vs direct Swiss Ephemeris
per-chart latency, hosted call vs local math: 1-30 s → ms; per-chart latency, hosted call vs local math

Read the case study Private repo Live site

Delivery with coding agents October 2026

Muse: Content-Scouting Agents on a Claude Subscription

A scheduled multi-agent pipeline that reads before it recommends, with no API key and no public surface.

A personal agent pipeline that searches YouTube and the web, reads or transcribes each candidate before recommending it, writes Markdown notes to a knowledge base, and serves a private feed. Every model call goes through the headless Claude CLI on an existing subscription, with no API key.

candidates gathered on the first run (8 read, 7 queued): 66; candidates gathered on the first run (8 read, 7 queued)
tokens of fixed overhead per model call: ~2.5k; tokens of fixed overhead per model call
items queued by the verified scheduled production run: 7; items queued by the verified scheduled production run
first run, first commit, and verified scheduled production: Same day; first run, first commit, and verified scheduled production

Read the case study Unpublished

Delivery with coding agents Evaluation & test infrastructure May to October 2026

J.T. Worldteacher: X Spaces to Podcast Pipeline

From a live X Space to a listed podcast episode with show notes, using local tools and a free media tier.

A pipeline and Astro site that turn J.T. Worldteacher's recorded X Spaces into a published podcast: download, local Whisper transcription, agent-written show notes, Cloudflare R2 audio hosting, and a validated RSS feed deployed to GitHub Pages.

episodes published: 24; episodes published
of audio transcribed and hosted: ~66 h; of audio transcribed and hosted
from git push to live episode in the feed: ~1 min; from git push to live episode in the feed
commits with a Claude co-author trailer: 15 / 24; commits with a Claude co-author trailer

Read the case study Code Live site

Delivery with coding agents Evaluation & test infrastructure July 2026

Thrive Beyond Coaching: Capped Installments with Stripe

A Stripe subscription that stops itself after exactly three payments, with no database.

Online payments for a small coaching practice: server-created Stripe Checkout with prices locked to a server catalog, plus a 12-week program payable monthly whose subscription automatically stops after exactly three charges.

charges, then the subscription cancels itself: 3; charges, then the subscription cancels itself
unit tests across checkout and the installment cap: 15; unit tests across checkout and the installment cap
real bugs caught by agent review before release: 2; real bugs caught by agent review before release
from first checkout commit to reviewed, documented release: ~2 h; from first checkout commit to reviewed, documented release

Read the case study Private repo Live site

More Projects

Products, coursework, and experiments

Astrology Insights

Active

The live product behind three of the case studies above: a web application for astrological birth charts with AI-written interpretations, chat, and synastry reports. Built with Flask and PostgreSQL, with charts computed locally using Swiss Ephemeris.

Python Flask PostgreSQL Swiss Ephemeris

Live site

AI Agent Orchestration

Active

Building and orchestrating teams of AI coding agents using Claude Code, OpenClaw, Codex, and Paperclip. Using agents to autonomously build features, write content, fix bugs, and run overnight — across multiple live projects simultaneously. Learning what works, what does not, and how to get the most out of AI-assisted development in real production environments.

Claude Code OpenClaw Codex Paperclip

Article

Embedded Log Anomaly Detection

Complete

A machine learning pipeline for prioritizing anomalous embedded-system test logs using Sentence Transformer embeddings and cosine similarity. Built for a CS 589 project, it compares several BERT-based encoders, normalizes timestamp-heavy data, and fine-tunes all-MiniLM-L6-v2 to separate abnormal and normal test runs without exposing proprietary log content.

Python Sentence Transformers BERT Machine Learning

Article

LSTM Stock Price Prediction

Complete

A neural network-based stock price prediction system using Long Short-Term Memory (LSTM) networks. Built as a final project for CS 682 Neural Networks course at UMass Amherst. Implements a multi-layer LSTM architecture with dropout regularization, layer normalization, and early stopping for predicting daily stock closing prices.

Python TensorFlow Keras LSTM

Code Paper

mattmccarthy.dev

Active

A modern, responsive personal website built with Flask and Tailwind CSS featuring blog, projects showcase, and interactive resume.

Python Flask Tailwind CSS Web Development

Code Live site

Interested in Collaborating?

I'm always open to discussing new projects, creative ideas, or opportunities to be part of your visions.

Get In Touch