brojonat
- 42 skills
- 0 followers
- 6 hours ago last updated
- ▌ Sqlite · brojonat bundleExtend SQLite with application-defined functions and loadable extensions in Python. Use when you need custom SQL logic (transforms, scoring, aggregations) in-process, or need to load/build C extensions for reusable SQLite plugins. Assumes Python stdlib sqlite3. Go equivalents exist (zombiezen/go-sqlite, mattn/go-sqlite3) — look up their APIs at runtime.
- ▌ Ibis Data · brojonat bundleUse Ibis for database-agnostic data access in Python. Use when writing data queries, connecting to databases (DuckDB, PostgreSQL, SQLite), or building portable data pipelines that should work across backends.
- ▌ Go Service · brojonat bundleBuild Go microservices with stdlib HTTP handlers, sqlc, urfave/cli, and slog. Use when creating or modifying a Go HTTP server, adding routes, middleware, database queries, or CLI commands.
- ▌ Scikit Learn · brojonat bundleBuild ML pipelines with scikit-learn, including preprocessing, cross-validation, hyperparameter tuning, evaluation, and MLflow tracking. Use when training models, building pipelines, or running ML experiments.
- ▌ K8S Deployment · brojonat bundleDeploy services to Kubernetes with Docker multi-stage builds, kustomize overlays, and Makefile automation. Use when creating Dockerfiles, writing k8s manifests, or setting up deployment pipelines.
- ▌ Fastapi Service · brojonat bundleBuild FastAPI services with JWT auth, structlog, and Prometheus metrics. Use when creating or modifying a Python HTTP server, adding authentication, structured logging, or instrumentation to a FastAPI app.
- ▌ Temporal Python · brojonat bundleBuild Temporal applications in Python using the temporalio SDK. Use when creating workflows, activities, workers, clients, signals, queries, updates, child workflows, timers, retry policies, saga/compensation patterns, testing, or any durable execution pattern in Python.
- ▌ Sklearn Pipeline Template · brojonat bundleBuild a tabular ML project with scikit-learn Pipelines, MLflow tracking, model artifacts, and a marimo demo notebook. Use when starting any new tabular classification or regression bundle so all bundles share the same plumbing.
- ▌ Building Bundles · brojonatHow to design, build, validate, and ship a new ManagerPack skill bundle. Use when starting any new bundle in the studio. Captures the workflow from picking a model family through to a published, marimo-validated, MLflow-tracked deliverable.
- ▌ Pymc Inference Template · brojonat bundleBuild a Bayesian inference project with PyMC, NUTS sampling, ArviZ diagnostics, MLflow tracking, idata persistence, and a marimo demo notebook. Use when starting any new Bayesian bundle so all bundles share the same plumbing.
- ▌ Air Go · brojonatHot-reload Go apps with cosmtrek/air during development. Use when setting up dev workflows for Go HTTP servers, configuring .air.toml, or debugging hot-reload issues with SQLite, port binding, or file watchers.
- ▌ Honker · brojonatAdd cross-process pub/sub, durable task queues, event streams, and scheduled jobs to a SQLite database using the honker extension. Use when you need Postgres NOTIFY/LISTEN semantics, background job processing, or event-driven architecture without leaving SQLite. Supports Python, Node, Go, Rust, Ruby, Bun, and Elixir via language bindings, or raw SQL via the loadable extension.
- ▌ Datastar · brojonatBuild real-time hypermedia web apps with Datastar — the backend owns state and pushes HTML down one long-lived SSE stream while writes are short-lived requests that render nothing. Use when working with `data-*` attributes, `datastar-patch-elements` / `datastar-patch-signals` SSE events, signals, `@get`/`@post` actions, or the `datastar-py` / `datastar-go` SDKs. Also use when a task calls for live-updating or multiplayer UI without a SPA.
- ▌ Ducklake · brojonatWork with DuckLake, an open lakehouse format built on DuckDB. Use when creating or querying DuckLake tables, managing snapshots, time travel, schema evolution, partitioning, or lakehouse maintenance operations.
- ▌ Bubbletea · brojonatBuild terminal user interfaces (TUIs) in Go with the Charm Bubble Tea framework (Elm-style Model/Update/View). Use when creating or modifying interactive CLI apps, dashboards, wizards, pickers, progress UIs, pagers, or any Go program that uses github.com/charmbracelet/bubbletea, github.com/charmbracelet/bubbles, or github.com/charmbracelet/lipgloss. Triggers on tasks involving tea.Model, tea.Cmd, tea.Msg, tea.KeyMsg, spinners, text inputs, lists, tables, viewports, progress bars, paginators, or styling terminal output.
- ▌ Python CLI · brojonatBuild Python CLIs with Click using subcommand groups. Use when creating or modifying a Python command-line interface, adding subcommands, or structuring a CLI package.
- ▌ Redfin API · brojonatInteract with Redfin's unofficial stingray API for property search, listing data, AVM estimates, and property history. Use when writing code to search listings, fetch property details, parse Redfin responses, or build GISCSV queries.
- ▌ Regression · brojonat bundleBuild a production-ready regression model on tabular data using XGBoost with conformalized quantile regression for prediction intervals. Use when the user needs to predict a continuous target from tabular features (price, sales, demand, time-to-event, score) and report uncertainty alongside the point estimate. Default to this for any tabular regression task.
- ▌ Urfave CLI · brojonatBuild Go command-line applications with urfave/cli v3 — commands, subcommands, flags, env var binding, Before/After hooks, shell completion, and testing. Use when creating or modifying a Go CLI entry point (main.go), adding subcommands, wiring flags to environment variables, customizing help output, or testing CLI behavior. Covers the cli.Command declarative API, flag types (StringFlag, IntFlag, BoolFlag, DurationFlag, StringSliceFlag, TimestampFlag, etc.), cli.Exit exit codes, and context.Context propagation into actions.
- ▌ Tabular Eda · brojonat bundleProfile a new tabular dataset before modeling. Find target leakage, missing data patterns, high-cardinality categoricals, near-constant features, redundant pairs, and non-linear relationships that Pearson correlation misses. Use whenever the user hands you a CSV or parquet and asks "what should I do with this?" Always run this skill before training any model on data you haven't seen before.
- ▌ Temporal Go · brojonatBuild Temporal workflow applications in Go. Use when creating or modifying Temporal workflows, activities, workers, clients, signals, queries, updates, retry policies, saga patterns, or writing Temporal tests.
- ▌ Pg Messaging · brojonatBuild pub/sub and task queue services using PostgreSQL as the backend. Use when implementing messaging primitives, log-based consumer groups, job queues with SKIP LOCKED, or replacing external message brokers (Kafka, RabbitMQ, SQS) with Postgres.
- ▌ Resend Email · brojonatSend transactional emails via Resend using SMTP or HTTP API. Use when integrating Resend for magic links, scheduled emails, cancellation, or listing sent emails. Covers API keys, SMTP config, scheduled_at, and Go integration patterns.
- ▌ Unsupervised · brojonat bundleFind structure in unlabeled tabular data — clustering with KMeans / GMM / DBSCAN and proper K selection, IsolationForest anomaly detection, and PCA dimensionality reduction. Use when the user has tabular data without a target column and wants to discover segments, find anomalies, or reduce dimensions. Always run tabular-eda first to profile the data.
- ▌ Data Flywheel · brojonatBuild a data flywheel service that ingests labeled data into SQLite, triggers async model training via honker queues after every N records, and hot-swaps the active model for inference. Use when you need a self-improving ML service backed by a single SQLite database with no external broker or scheduler.
- ▌ Openai Agents · brojonatBuild multi-agent systems with the OpenAI Agents SDK, including tool definitions, handoffs, context management, and webhook validation. Use when creating OpenAI agent flows, defining tools, or handling agent handoffs in Python or Go.
- ▌ Litestream K8S · brojonatRun SQLite with Litestream replication to S3-compatible storage (Cloudflare R2, etc.) on Kubernetes. Use when adding persistent SQLite to a containerized app, setting up Litestream, or doing point-in-time database recovery.
- ▌ LLM Finetuning · brojonat bundleFine-tune small open-weight LLMs locally with Unsloth + QLoRA for text tasks (classification, extraction, translation). Use when input is raw text (not tabular features), user has labeled examples, and wants model ownership without API costs. Default to Gemma-4 E2B; scale up based on VRAM. Always compare zero-shot vs fine-tuned, log every run to MLflow, export to GGUF for llama.cpp.
- ▌ Playwright E2e · brojonatWrite Playwright end-to-end tests for web apps. Use when setting up e2e tests, testing auth flows, or writing browser automation scripts. Covers project setup, test patterns, auth bypass for dev mode, and integration with Go/backend servers.
- ▌ Marimo Notebook · brojonatAuthor marimo notebooks with conventions for interactive scatter exploration. Use when creating a new marimo notebook, choosing how to run marimo locally, or wiring up jscatter scatter plots with lasso selection feeding a table below the plot.
- ▌ Bayesian Bandits · brojonat bundleImplement Thompson sampling for multi-armed and contextual bandits. Use when the user wants to adaptively allocate traffic across variants (ads, recommendations, content, pricing) to minimize regret instead of running a fixed-allocation A/B test. Covers Bernoulli bandits, contextual bandits, regret analysis, and comparison with epsilon-greedy and UCB.
- ▌ Data Labeling QA · brojonat bundleAudit an untrusted labeled training set before fine-tuning. Find mislabeled examples with four complementary techniques — provenance/integrity checks (off-by-one, schema drift), cleanlab confident learning (random noise), high-loss monitoring (hard cases), and LLM-as-judge on the flagged subset (systematic confusion). Use before spending GPU on any fine-tune where you don't fully trust the labelers. Works with any classification dataset reshaped to `data` and `untrustworthy_label` columns.
- ▌ Parquet Analysis · brojonat bundleAnalyze parquet files using Python and Ibis. Use when the user wants to explore, transform, or analyze parquet data files, perform aggregations, joins, or export results. Works with local parquet files and provides database-agnostic data operations.
- ▌ Pyproject Config · brojonatConfigure Python projects with pyproject.toml including build system, console scripts, ruff linting, and pytest. Use when setting up a new Python project, configuring tooling, or adding entry points.
- ▌ Bayesian Ab Testing · brojonat bundleRun a Bayesian A/B test on conversion data using PyMC. Use when the user wants to compare two variants (landing pages, emails, pricing, UI changes) and decide which to ship using posterior probabilities and expected loss instead of p-values. Covers Beta-Binomial model, ROPE, expected loss, sample-size guidance, and ArviZ diagnostics.
- ▌ Bayesian Regression · brojonat bundleFit Bayesian regression models with PyMC using the Hogg approach — start simple, diagnose problems, upgrade the likelihood. Use when the user needs regression with proper uncertainty quantification, heteroscedastic errors, outlier robustness, or model comparison. Covers Normal, Student-t, and GLM likelihoods with ArviZ diagnostics and LOO-CV.
- ▌ Binary Classification · brojonat bundleBuild a production-ready binary classifier on tabular data using XGBoost. Use when the user needs to predict a binary outcome from tabular features (churn, fraud, conversion, default, click). Covers class imbalance, threshold tuning, calibration verification, and SHAP feature importance. Default to this for any binary classification task on tabular data.
- ▌ Bayesian Mixture Models · brojonat bundleFit Bayesian Gaussian mixture models with PyMC for soft clustering with full uncertainty. Use when the user needs probabilistic cluster assignments, latent heterogeneity modeling, zero-inflated data, or principled model selection for number of clusters. Covers Dirichlet priors, label switching, LOO-CV model comparison, and applications beyond clustering.
- ▌ Multiclass Classification · brojonat bundleBuild a production-ready multiclass classifier on tabular data using XGBoost. Use when the user needs to predict one of several discrete classes from tabular features (product category, sentiment level, customer segment, intent, fault type). Covers per-class metrics, confusion matrix analysis, sample weighting for imbalance, top-K accuracy, and SHAP. Default to this for any tabular multiclass problem.
- ▌ Multilabel Classification · brojonat bundleBuild a production-ready multilabel classifier on tabular data using XGBoost wrapped in MultiOutputClassifier. Use when each row can have multiple labels simultaneously (tags, attributes, gene functions, content moderation categories, multi-disease detection). Covers hamming loss, per-label metrics, label co-occurrence, MultiOutputClassifier vs ClassifierChain, and per-label SHAP. Default to this for any tabular multilabel problem.
- ▌ Bayesian Decision Analysis · brojonat bundleTurn posterior distributions into optimal actions using loss functions, expected value of information, and the newsvendor framework. Use when the user has a Bayesian model and needs to make a decision under uncertainty — pricing, inventory, ship-or-wait, resource allocation. Covers EVPI, EVSI, custom loss functions, and asymmetric cost structures.
- ▌ Statistical Process Control · brojonat bundleMonitor a metric over time with process behaviour charts (XmR / control charts) to separate routine variation from real process change. Use whenever the user wants to know "is this number unusual?", set alerting thresholds on a KPI or operational metric, detect anomalies or regressions in a time series of measurements, build a control chart, or replace ad-hoc "it looks high this week" judgments with something principled. Default to this for metric monitoring and threshold-setting tasks — even when the user doesn't say "SPC" or "control chart" — including dashboards, SLO/error-rate watching, manufacturing quality data, and weekly business reviews.