The tech stack behind Growth Bastards

01Languages and runtime

Python 3.12+: Backend services and analysis; Python 3.13 on the development Mac.
TypeScript and Node.js 22: Web applications and Dr. UX workers.

Anthropic Claude: Model calls through the Python 1.x and JavaScript SDKs; Opus and Sonnet routing.
Voyage AI: Text embeddings.
Amazon Bedrock and boto3: Secondary embedding route; image and document reading.
OpenAI, Google Gemini and Exa: External services used by Boswell.
YouTube and youtube-transcript-api: Transcript collection.

PostgreSQL 16: Relational database; local instances for development and tests.
Neon: Managed AtmaOS database; protected main branch and seven-day point-in-time recovery.
Supabase: Database, authentication and server rendering helpers for Assumption Annihilator.
pgvector and pg_trgm: Vector search and text similarity; PostgreSQL also supplies full-text search.
psycopg 3 and pg: Python and Node.js database access.
SQLAlchemy and Alembic: Database access and migrations for the compliance engine.
AtmaOS SQL migration runner: Checksummed, append-only migrations.
PostgreSQL access controls: Database roles, row-level rules, triggers and signature guards.
Object storage: AtmaOS file storage, alongside Cloudflare R2.

Starlette, FastAPI and Uvicorn: HTTP APIs and API hosting.
httpx and requests: Outbound HTTP calls.
Meta Marketing API v26, Google Ads API and LinkedIn Marketing API: Advertising connectors.
Google Analytics 4 and Shopify exports: Analytics and order data.

React 18, Next.js and Vite: Web applications and frontend builds.
Tailwind CSS and Radix UI: Styling and interface components.
Framer Motion, Recharts and React Flow: Animation, charts and interactive diagrams.
oidc-client-ts: OpenID Connect sign-in.

NumPy, SciPy and pandas: Numerical analysis and data preparation.
JAX, jaxlib and NumPyro: Bayesian computation and inference.
ArviZ: Posterior analysis and inference diagnostics.
PyArrow: Columnar data processing.
scikit-learn: Classification, regression, clustering and model evaluation.
XGBoost; CatBoost alternative: Prediction from structured customer and campaign data.
BERTopic, HDBSCAN and optional UMAP: Topic discovery and clustering.
Sentence Transformers: Local text embeddings and retrieval reranking.
SHAP: Explanations of model predictions.
statsmodels and scikit-survival: Forecasting and time-to-outcome analysis.
EconML: Causal forests and double machine learning.

Apache Parquet: Versioned analytical datasets and training snapshots.
DuckDB and isolated Python workers: Analytical SQL, feature preparation, training and inference.
Apache Iceberg and Apache Polarisconditional: Shared analytical tables, snapshots and concurrent writes.
Trinoconditional: Distributed analytical SQL when single-node benchmarks fail.

AtmaOS metric service: Shared metric definitions and authorized queries.
MetricFlow and dbt Semantic Layerconditional: Central metric calculation and join rules.
Hybrid retrieval and Reciprocal Rank Fusion: Combined keyword and vector evidence search.
HNSW indexing and Qdrantconditional: Faster vector retrieval when measured targets require it.
AtmaOS feature and model records: Feature definitions, model versions, evaluations and deployment history.
MLflowconditional: Shared model registry and release interface.
Feastconditional: Shared, current features for online predictions.
OpenLineageconditional: Data and model lineage across transformation systems.

PostgreSQL job queue and transactional outbox: Durable analytical jobs, checkpoints and retries.
Dataset publishing workers: Validated, versioned datasets and artifact manifests.
Arkwright dashboard and engine adapters: Evidence review, specialist collaboration and authorized action requests.

python-docx, lxml, olefile and mammoth: Word and legacy Word processing; XML handling.
pdfplumber, pypdfium2 and unpdf: PDF reading and extraction.
Pillow: Image processing.
ReportLab and Jinja2: PDF generation and report templates.
openpyxl: Excel reading and writing.

Pydantic 2, pydantic-settings and Zod: Record validation, data contracts and typed settings.
PyYAML and YAML registries: Identity files, statistical defaults, diagnostics and rule configuration.
python-dotenv: Local environment configuration.
Compliance rule packs: Claim and rule review definitions.

GitHub and GitHub Actions: Source control and continuous integration.
Docker Compose: Local service environments, including databases and Dr. UX egress.
pnpm and tsx: JavaScript workspace management and TypeScript scripts.
Make and deployment scripts: Build and deployment commands.
Vercel: Assumption Annihilator hosting.