Projects

A selection of systems designed and built by Quant-ED partners.

District Haystack (2024 – present)

A Django REST Framework platform that orchestrates a distributed crawl of K-12 school district websites and makes the results searchable. A Celery/Redis async pipeline handles document extraction and embedding; Docling extraction runs with memory-pressure backoff and bounded worker pools, with legacy Office conversion handled by a LibreOffice sidecar. VoyageAI embeddings power semantic search over an authenticated API, on a PostgreSQL/Nginx/Docker stack served over a private Tailscale network.

Distbot (2024 - present)

A fully async FastAPI crawler microservice built on Crawlee — the collection engine behind District Haystack. It switches between a Camoufox browser backend for JavaScript-heavy sites and curl-impersonate for static ones, prioritizes URLs likely to contain policy content, refreshes tiered proxies automatically, and offers pause/cancel/reset job control with resource limits for parallel district crawls.

ELI Reflect (2026)

A Django platform collecting video, audio, and text reflections from educators in Hawaiʻi. Built for real classroom conditions: browser-based media capture with an IndexedDB offline retry buffer, Google OAuth2 sign-in, Google Drive storage, and speech-to-text with automatic English/Hawaiian language-switch detection plus Claude-powered translation. Provisioned with Ansible and deployed via Podman with TLS.

Research Hub (2026)

A self-hosted retrieval-augmented generation application for research literature. PDFs are deduplicated by content hash and extracted with Docling into reusable JSON sidecars; OpenAI embeddings in ChromaDB ground a fully async, SSE-streamed chat that cites its document sources. Django 5.1, Celery, PostgreSQL.

District Fuzzyfinder (2024 – 2025)

A fast fuzzy-search service for school district IDs, names, and websites drawn from NCES Common Core Data — the small piece of glue that keeps every other education dataset joinable.

Partner experience

Beyond Quant-ED engagements, our partners bring production experience from the University of Washington, including:

  • A nine-state education labor-market scraping platform: ~100 independent district-site scrapers delivering ~100,000 structured job-posting records weekly, with 94–100% coverage in most states.
  • An LLM policy-analysis platform scaled to 8,000+ policy documents and 414,000+ structured AI queries, powering published research on school district grading policies and teacher collective bargaining agreements.
  • A decade of education policy research management: dozens of peer-reviewed publications on teacher labor markets, school staffing, and pandemic-era schooling, and analysis supporting state boards of education and legislative testimony.

Get in touch if you'd like to build something similar.