Machine learning

Build a repeatable eval loop that grades agent output with an LLM judge, so prompt/skill changes get scored against a baseline instead of eyeballed.

Use when fine-tuning LLMs, training custom models, or adapting foundation models for specific tasks.

Designs and implements production-grade ML pipeline infrastructure: configures experiment tracking with MLflow or Weights & Biases, creates Kubeflow or…

Writes, refactors, and evaluates prompts for LLMs — generating optimized prompt templates, structured output schemas, evaluation rubrics, and test suites.

Designs and implements production-grade RAG systems by chunking documents, generating embeddings, configuring vector stores, building hybrid search pipelines…

Validate and use packed sequences and long-context training in Megatron-Bridge, including offline LLM packing, collate-time VLM packing, Energon online…

AI/LLM application security testing — prompt injection, jailbreaking, data exfiltration, and insecure output handling per OWASP LLM Top 10.

A specialized skill for embedding and extracting resilient watermarks in text by manipulating sentence lengths and using Fountain Codes.

Uses the Biohub esm Python SDK for ESM3 protein generation, ESMC embeddings, and ESMFold2 all-atom folding.

Extracts and preprocesses whole-slide histology image tiles with Histolab.

Plans and audits use of ChicagoHAI HypoGeniC/HypoRefine for LLM-assisted hypothesis generation from labeled text datasets.

Featurizes small molecules with Molfeat for QSAR/QSPR, chemical similarity, virtual screening, and molecular ML.

Supports local computational pathology research with PathML: slide loading and tiling, preprocessing and QC, h5path storage, multiplex quantification, spatial…

Builds and checks Bayesian models with PyMC, including hierarchical models, NUTS MCMC, variational inference, mutable-data predictions, posterior predictive…

Hugging Face Transformers for loading Hub models, running pipeline inference, text generation, and Trainer fine-tuning on NLP, vision, audio, and multimodal…

Generates Python code using BigQuery DataFrames (BigFrames).

Summarize last 7 days of compass drift events · alert distribution + top 5 negative drifts · pure read-only · NO LLM.

Scan project memory · age distribution + ungrouped count (L1 build readiness) + per-namespace entity stats · NO LLM.

Use parallel subagents to scale large reviews, with an honest model of where the speedup actually comes from (LLM reasoning, not API throughput)

Convert `make demo/img` statusline snapshots into ANSI-stripped plain text for diffing and PR embedding.

Autonomously optimizes end-to-end LLM inference throughput on AMD Instinct GPUs and reports a validated gain, using the Hyperloom multi-agent optimizer.

Integrates local AI capabilities into applications using Embeddable Lemonade.

Benchmarks LLM inference and drives GPU kernel optimization with Magpie.

Diagnoses why ROCm, the HIP SDK, PyTorch, or llama.cpp is broken on an AMD GPU on Linux, Windows, or WSL2, then applies a low-risk fix with consent or hands…