Nelson Love — Applied AI · Test Automation · Developer Tooling
Production LLM systems and test automation for regulated software.
Applied-AI and tooling engineer for clinical, behavioral-health, and life-sciences platforms — LLM pipelines, structured extraction, and the test infrastructure that keeps them honest; background in psychology research and FDA-regulated biotech (21 CFR Part 11, GAMP).
Open to new roles · remote (US)
Profile
Background
I started in psychology research — designing studies, running the statistics, publishing on how people find decent work — and moved into software the same way I move through any system: learn how it actually behaves, then build the tooling to keep it honest.
For two and a half years I was an SDET at L7 Informatics, building test automation and CI/CD infrastructure for a LIMS platform used in FDA-regulated labs — working day to day inside 21 CFR Part 11 and GAMP. I also built the AI and developer tooling around it: an LLM structured-extraction pipeline that turned thousands of opaque test failures into queryable data, an embeddings-plus-GPT system that audited 7,750 test-to-requirement mappings with human-in-the-loop evaluation, and a reviewer-recommendation engine where a small, legible algorithm beat the obvious ML approach.
Since then I've worked independently — building LLM-powered developer and go-to-market tooling, ETL pipelines over FDA compliance data, a live audio-processing SaaS operated end to end in production, and MCP infrastructure for AI agents.
What ties it together is a bias toward systems that are both rigorous and humane: software that's correct under audit and clear to the people using it. That instinct is sharpest in clinical, behavioral-health, and life-sciences software — where correctness and human stakes are the same thing.
Selected work
7 entries
-
L7 Informatics — SDET
- Python
- LLM function-calling
- structured output
- test infra
Turned thousands of opaque Cypress failures into structured, queryable data using LLM function-calling — a structured-extraction pattern built before it was standard.
-
L7 Informatics — SDET
- Python
- embeddings
- LLM evaluation
- 21 CFR Part 11
A two-stage embeddings-plus-GPT pipeline that audited 7,750 test-to-requirement mappings for regulated traceability, with human-in-the-loop validation, caching, and cost controls.
-
L7 Informatics — SDET
- Python
- recommendation system
- git analysis
- Bitbucket API
Routed PRs to the right reviewers via git-history analysis with exponential-decay weighting — a legible algorithm that beat the obvious ML approach.
-
Independent contractor
- Python
- OpenAI API
- ETL
- SQLite
- data pipeline
A Python tool that profiles manufacturers against a 10,000+ record FDA-compliance database, classifies them with the OpenAI API, and drafts targeted outreach.
-
Independent — currently building
- MCP
- Python
- retrieval
- agent tooling
A hosted Model Context Protocol server that safely exposes a 7,000-note knowledge base to AI agents, backed by a typed information model and a SQLite store.
-
Independent venture
- FastAPI
- Vue
- Docker
- Terraform
- RunPod
A live audio-processing SaaS — FastAPI and Vue on Docker with Terraform-provisioned GPU inference on RunPod — built and operated end to end.
-
Steady - pacing and symptom tracking for chronic illness
In developmentIndependent venture
- product
- health
- forecasting
- measurement
A measurement-driven app for people managing ME/CFS, long COVID, and POTS — pacing and crash forecasting from time-series symptom data, grounded in real psychometrics.
Projects
18 entries →-
Logit
Jul 2026- psychometrics
- rasch
- next.js
- fastapi
- R
- supabase
A browser-based Rasch measurement workbench that calibrates test and survey data, produces the full diagnostic suite, and checks every estimate against five published reference calibrations, as a modern alternative to Windows-only desktop software.
-
Balance of Power
Jul 2026- python
- textual
- simulation
- games
Source-faithful Python port of Chris Crawford's 1990 Cold War simulation — a headless engine plus a full terminal UI, playable by humans and agents alike.
-
Thinkblot
Jul 2026- python
- fastapi
- react
- psychology
Rorschach Comprehensive System scoring platform — a Python scoring engine, a stateless FastAPI service, and a React scoring workbench for clinicians.
-
safari-py
Jul 2026- python
- macos
- cli
Python library and CLI for Safari bookmarks and the reading list on macOS — read, search, organize, and export, with JSON output and dry-run on every write.
-
vault-skills
Jul 2026- obsidian
- claude code
- agents
- typescript
An Obsidian plugin that compiles vault notes into a native Claude Code plugin — skills, agents, and policies authored as markdown, validated as a tree, exported as working agent infrastructure.
-
Claude Code Plugins
Jun 2026- python
- claude-code
- macos
A marketplace of Claude Code plugins wiring the assistant into native macOS apps.
-
Dotfiles
Jun 2026- shell
- macos
- dotfiles
My macOS dotfiles and bootstrap — shell, editor, terminal, git, and tmux.
-
Obsidian Vault MCP Server
Jun 2026- typescript
- mcp
- obsidian
A remote MCP server that exposes an Obsidian vault folder to Claude.
-
JD Obsidian
May 2026- typescript
- obsidian
- johnny-decimal
A Johnny Decimal dashboard plugin for Obsidian.
-
Claude Config GUI
May 2026- swift
- macos
- claude-code
A native macOS app for managing Claude Code configuration.
-
Exnerator
Aug 2021- python
- psychology
Computerized scoring for the Rorschach Comprehensive System
-
Coding Exercises
Aug 2021- python
- lisp
Solutions in Python and LISP to Codewars and Project Euler problems
-
gpt-utils
Aug 2021- python
- gpt-3
Helper tools for use with OpenAI's GPT-3 API
-
Tock
Aug 2021- php
Time tracking and scheduling app written in PHP
-
Animal Facts
Aug 2021- python
- flask
- gpt-3
A tiny Flask-based "meme generator".
-
Taipan
Aug 2021- python
- games
Python port of the classic Apple II game
-
- astro
- web
This website. Built with Astro.
-
- python
- nlp
- llm
- machine-learning
- life-sciences
- qa
Two-stage NLP/LLM pipeline that evaluated 7,750 test-requirement mappings for a regulated life sciences platform, reducing a projected 3–4 week manual effort by six engineers to roughly two weeks.
Writing
9 posts
-
Notes from the two years before ChatGPT — IQ-testing davinci at 2 a.m., whale facts, an Abe Lincoln chatbot, and a machine that told me it worried it was a toy.
-
A sonnet generator I remembered from a 1995 Mac book turned out to be uncatalogued anywhere online — so I dug it out of my own disk image, decoded how it works, and put it back on the internet twice.
-
Notes on a compliance-data pipeline — scraping, normalizing, and scoring 10,000+ establishments.
-
How nelson.love publishes itself from an Obsidian vault, and what it was like to build that with an AI agent reading the notes over your shoulder.
-
What the reviewer-recommendation engine taught me about reaching for the legible solution first.
-
Using models to turn messy test output into clean, queryable data — the boring, useful application.
-
Why you have five note apps: a three-axis model — object type × verb × register — of the whole space personal-information tools occupy, and the orchestration layer that belongs between them.
-
The most reliable place to put a model is between two systems that already know what they want.
-
AI coding agents fail when requirements underdetermine the domain model. A case for a machine-readable domain layer — an intermediate representation between domain knowledge and generated code.
Essays
9 essays →-
Children's stories and the love that makes you real — the Velveteen Rabbit, the fox, and Winnicott.
-
Levinas's ethics of the face confronted with GAN-generated faces — what the perfect idol reveals, harvests, and cannot counterfeit.
-
On the mental map, starlight, mediated catastrophe, and why we never quite live in the present.
-
Global warming as Teilhard's biosphere-to-noosphere threshold — and why the threshold is responsibility, not awareness.
-
On the anglerfish's lure and Icarus — being drawn toward something with no reason to be drawn, and toward the thing that undoes you.
-
Nietzsche's "God is dead" was a warning, not a demolition — and we are making the same misreading with Foucault.
-
Lacan's three registers are laid down one over another — and you learn what each one was doing by watching it fail.