How Good Deeds Dallas is built

Volunteer listings in this metro are scattered. A nonprofit in Plano might post to Idealist; one in McKinney posts to its city's Galaxy Digital portal; a third posts nowhere but its own website. Each portal uses its own category system, so “food security” work is filed under four different labels depending on where you look, and there is no single place to ask “what needs doing near me this weekend?”

This site aggregates those sources, applies one consistent set of cause tags, removes what has expired or was never a volunteer role to begin with, and links you straight back to the organization to sign up.

Six public sources are rechecked every week:

• Idealist — the largest national listing site, filtered to the DFW metro.
• Volunteer McKinney and Volunteer Garland — municipal Galaxy Digital portals with structured feeds.
• Voly Dallas — a Texas volunteer platform used widely by local nonprofits.
• Dallas Doing Good — local coverage of nonprofits and the volunteer needs they publish.
• Nonprofits' own websites — a curated list of local organizations that publish no structured feed at all. Their volunteer pages are read and converted into structured listings by a language model.

Local subreddits are read too, but kept separate as community chatter rather than mixed into the opportunity index.

Aggregation without filtering produces a worse experience than the portals it aggregates. Every extracted listing passes three stages before it appears:

• De-duplication. Platforms re-post the same shift many times — the same Saturday pantry slot can appear a dozen times under different dates. Rule-based matching collapses those.
• Expiry. One-time events drop off automatically the day after they happen. This runs daily, not weekly, so nothing shown here has already passed.
• Screening. A language-model judge removes what isn't actually a volunteer role: 5K race signups, donation drives, paid internships posted to the wrong board. Sources that are already staffed volunteer platforms skip this stage — they don't need it.

Rejected listings are kept in a separate audit file rather than silently dropped, so a filter that starts making bad calls is visible rather than invisible.

Each source labels causes its own way, which makes filtering across them meaningless. Every listing is re-tagged against a single fixed taxonomy of 18 categories — seniors, food security, animals, environment, mental health, and the rest. That is what makes it possible to ask for “animal welfare volunteering in Denton” and get results from four different portals in one list.

Alongside keyword search there's a retrieval-based one. Listings are embedded into a vector index; a query is embedded and matched by similarity; a language model then writes an answer using only the listings actually retrieved. It cannot invent an opportunity that isn't in the index — if nothing matches, it says so rather than inventing a plausible-sounding nonprofit.

The full collection, quality-control and re-tagging pipeline runs weekly and unattended. Expiry runs daily. The site rebuilds on every data change, so the counts on every page — 1,779 opportunities as of this build — are generated from the current data rather than written by hand.

Browsing requires no account and sets no tracking cookies. Smart Search is rate-limited per IP using a hashed address, and no raw addresses are stored. The full detail is in the privacy policy.