BioNodulo + AI evaluation Shipping since 2024

Pro jects

One big platform, a few smaller tools, and a side job grading AI agents on their coding.

Scroll
Fig. 01 — BioNodulo closed alpha · v0.1

Visual pipelines, node by node

A drag-and-drop canvas where every node wraps a real bioinformatics tool. Runs locally, on HPC, or on cloud. Exports to Nextflow, Snakemake, CWL and Galaxy.

943 nodes 51 categories 22 templates local · HPC · cloud
bionodulo.com
Node canvas placeholder
The canvascore

Every step visible. Wire tools together, see types and environments at each node, catch mismatches before you run.

AI agent placeholder
AI agentbuilt in

Describe the analysis and watch the graph assemble. It builds, validates, runs and fixes workflows, and reads the logs when something breaks.

DOI to pipeline placeholder
DOI to pipelineparty trick

Paste a paper's DOI. BioNodulo reads the methods section and suggests a pre-wired workflow that matches it.

Multiplayer placeholder
Multiplayer + MCPcollab

Edit the same canvas with your team in real time, or drive the whole editor from Claude, Codex or ChatGPT over MCP.

Fig. 02 — Grading the machines Dec 2025 – now

Between degrees I work as an AI evaluator, reading what coding agents produce and marking where they go wrong. It's the same job as the PhD, done by hand: catching plausible-but-wrong output.

2026– Mercor — AI Generalist ExpertProject Les Artistes Artifacts. Under NDA, so that’s all I can say. San Francisco, remote
2026 Alignerr — AI Systems EvaluatorTested Claude-based coding systems on real software tasks; structured preference feedback for Anthropic's coding models. Remote
2025–26 Alignerr — Agentic AI Code ReviewerReviewed Claude Code agent transcripts against a weakness taxonomy: hallucinations, overengineering, laziness, instruction-following failures. Remote
Fig. 03 — Also in the workshop
OSS Agentic AI harnessesOpen-source tooling for running complex bioinformatics workflows with agents. github.com/Classacre
ML Model trainingLoRAs, hypernetworks and image models, trained at home for fun. off hours
SITE This websiteHand-rolled HTML, CSS and JavaScript. Lenis smooth scroll, Three.js shaders, no framework. The source is public. view source
About The person behind the nodes ↗