0ee478ec3d
matchGrants (nightly 05:15): per candidate org (NH + good standing + primary ICP) ensure a v0 NTEE-derived profile embedding (RETRIEVAL_QUERY, confidence 0.2 stub the Stage 4 profiler upgrades in place), retrieve top-50 open grants by pgvector cosine with SQL-enforced cheap gates (deadline >=21d, ceiling >=10K), run entity/geography gates in TS, score deterministically, upsert pair-keyed matches, reassign hero. Scoring: pure scoreMatch (mission fit 30 / capacity 15 / competition 15 / effort 10 / runway 5; precedent's 25 reserved until the 990-PF index; easy win >= 50/75). Eligibility gate now pattern-matches Grants.gov applicantTypes prose, conservatively (ambiguous entries do not admit). application_form_supported ignored for pass/fail per the manual-first decision, still recorded in rationale. Schema: unique (org_id, grant_id) on matches; unique org_id on org_profiles (latest-profile semantics). Review queue query now ordered hero > easy-win > score and capped at 100. Live run: 64 orgs -> 3,200 matches in 28s, 0 easy wins / max 39 — the honest result of an NIH-heavy 200-grant corpus vs NH service nonprofits; engine mechanics verified, corpus breadth is the next lever. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
HelmDocs Grant-Match Outreach Engine
Automated pipeline that matches open grants to NH nonprofits, then runs a personalized email drip whose success metric is a booked demo call. The grant match is the lead magnet — match accuracy is the prime engineering directive.
Architecture doc: docs/plan.md. Build plan and phase status: ~/.claude/plans/helmdocs-outreach.md.
Relationship to novelpad-desktop
This repo deliberately mirrors the novelpad-desktop stack so a future merge is a directory move, not a rewrite:
- Yarn 4.5.0 + Turbo monorepo, Node 22.16 (volta), TypeScript ^5.9.3
- Postgres + pgvector + Drizzle (0.44.6, pinned via
resolutions), actions/queries pattern - DBOS for durable jobs/cron (worker-app pattern)
@google/genaion Vertex; embeddingsgemini-embedding-001@ 1536 dims (identical to HelmDocs RAG — never mix embedding models)- React Router v7 for the review UI
Merge rules: all packages stay in the @novelpad/ scope with names that never collide with novelpad-desktop workspaces (outreach-*). Files copied from novelpad-desktop carry a provenance header (source path @ commit) — diff against origin at merge time. No live code sharing; drift is fixed at merge, by design.
Boundary: separate repo and separate Postgres on purpose — outreach tooling and sourced contact data stay outside the compliance-scoped product repo/DB.
Workspaces
| Workspace | Purpose |
|---|---|
packages/config |
Shared tsconfig/tailwind base (copied from novelpad-desktop) |
packages/outreach-core |
Drizzle schema (grants, orgs, profiles, contacts, matches, pipeline_events), actions/queries, hard gates |
packages/outreach-ai |
Gemini client, embeddings, org-profiler + mission-fit-judge agents |
apps/outreach-worker |
DBOS executor: ingestion crons, expiry sweep, scoring runs, Apollo sync |
apps/outreach-review |
Human review queue (RR7). No auth by design — run behind IAP/localhost |
Development
yarn install
yarn build
yarn typecheck
yarn test
Migrations: yarn create:migration <name> inside packages/outreach-core (server-only — no client DB in this repo).
Local infrastructure
docker-compose.yml mirrors novelpad-desktop's compose conventions (pgvector image, healthcheck-gated startup), trimmed to this repo: Postgres 16 + pgvector on host port 54341 (novelpad's stacks use 54321/54331 — all three run side by side), plus worker and review services built from the root multi-stage Dockerfile (same image, different command — the novelpad website/auth pattern).
docker compose up -d postgres # just the database
DATABASE_URL=postgresql://postgres:password@localhost:54341/helmdocs_outreach \
yarn workspace @novelpad/outreach-core migrate
docker compose up -d # full stack (worker + review UI on :3100)
Migrations always run from the host — drizzle-kit is a devDependency and deliberately absent from the production image. The DBOS system schema initializes itself into the same database on worker boot (the dbos schema; ignore the misleading *_dbos_sys URL in DBOS's boot log — the shared pool override wins).