leadplatform — Repo Digest

The inbound side of Inexis lead generation: triage buyer-intent work opportunities and route each to the right business with a decision brief. Consumes the Intelligence Products packs; complementary to the Website Assessment engine.

What it is

“leadplatform — Lead / Opportunity Intelligence Platform. Internal decision-support tool for two Australian businesses sharing one platform. It triages inbound work opportunities: classify → score → route to the right business → produce an internal decision brief.” (CLAUDE.md). Explicitly “not a generic Upwork job scraper… a Lead Intelligence Platform that happens to use Upwork as its first live lead source.” (docs/vision).

Why it exists

Inbound opportunities (people actively looking to buy) are high-leverage but noisy. This platform qualifies and routes them deterministically to Inexis Digital (websites) or Inexis Consulting (AI/automation), producing a decision brief — so a human can bid with context. It “does not submit proposals, run outreach, or act as a CRM — humans still bid manually in Upwork.”

At a glance

Business capability provided

Opportunity intelligence & routing — classify, score, and route inbound buyer-intent jobs to the correct Inexis business, with a decision brief and a 0–100 Upwork opportunity score (70+ surfaces).

Technical responsibilities

• Own classification, scoring, routing, brief generation, and live Upwork ingestion → D1.
• Explicitly does not: “submit proposals, run outreach, or act as a CRM”; and per ADR-002 excludes automated proposal submission, outreach automation, and any change to the website-assessment engine.

Core concepts

Key workflows

1. Live ingestion (Worker): OAuth2 Upwork fetch → pipeline.ts (classify + route; no scoring yet) → db.ts upsert (de-dupe on id) into D1.
2. Engine validation: npm run validate:upwork over data/upwork-sample-jobs.json.
3. Interactive triage (Next.js): manual job list → /new → /[id] decision brief.

Technologies used

• Two runtimes, one repo, sharing pure lib/ engines:
  • Web: Next.js 16, React 19, shadcn/Tailwind v4, deployed via OpenNext → Cloudflare; localStorage persistence (Phase 1).
  • Worker: Cloudflare Workers + D1, TypeScript, wrangler ^4, cron */30.
• @base-ui/react, tsx for scripts. Two wrangler configs (leadplatform-web, leadplatform-ingest).
• “This is NOT the Next.js you know” (AGENTS.md) — Next 16 has breaking changes.

Major modules / components

Capabilities

• opportunity-intelligence — classify/score/route inbound opportunities
• lead-ingestion — live Upwork OAuth2 ingestion → D1
• decision-brief — internal brief generation

Upstream dependencies

Downstream consumers

• Human operators (internal decision-support) — no programmatic downstream consumer.
• Uses intelproducts skills/proposal-generator to draft proposals (proposals/).

Major interfaces & integration points

• Worker endpoints (leadplatform-ingest.*.workers.dev): GET /health, GET /jobs?agency=…, POST /ingest, GET /upwork/preview, POST /upwork/ingest, GET /upwork/whoami; cron */30 * * * *.
• Upwork auth: OAuth2 authorization-code + refresh token; secrets via wrangler secret put (UPWORK_CLIENT_ID/SECRET/REFRESH_TOKEN/ORG_ID).
• Pack contract: vendored intelproducts submodule (pin a version; treat as read-only).

Reusable assets exposed to other repositories

• The pure lib/ decision engines (runtime-agnostic; the Worker reuses the same engines as the web app).
• The routing table and the Upwork OAuth bootstrap pattern.

Architectural decisions

• [ADR-001] Phase-1 MVP scope (Accepted) — deterministic evaluation; exclusions.
• [ADR-002] Upwork as first live lead source (Accepted 2026-06-16) — consolidate on Workers + D1; scoring deferred. Plus docs/vision, docs/roadmap, docs/specs (classification, opportunity-scoring, knowledge-matching, opportunity-brief), docs/architecture/mvp-architecture.md.

Architecture snapshot

graph TD
    UP[Upwork GraphQL API] -->|OAuth2, cron */30| W[Cloudflare Worker<br/>worker/src/*]
    W --> LIB[Shared lib/ engines<br/>classify · route · score]
    LIB --> DB[(D1 · jobs)]
    WEB[Next.js 16 web app<br/>OpenNext → Cloudflare] --> LIB
    OP([Human operator]) --> WEB
    PACK[(intelproducts pack<br/>pinned submodule)] -.read-only.-> LIB

Current maturity

Active-development. Phase 1 (Next.js MVP) is “complete and validated”; Phase 2 decision logic is done (verified by npm run validate:upwork); the live Upwork client is implemented but token bootstrap is pending, and the Worker computes “no score yet” (scoring deferred per ADR-002). Phases 3–6 (intelligence extraction, proposal assistant, CRM + outcome learning, pursuit automation) are planned.

Roadmap

Phase 3 AI qualification / intelligence extraction → Phase 4 Proposal Assistant (demo selection) → Phase 5 CRM + outcome learning → Phase 6 pursuit automation. “Rule-based now, AI-swappable later.”

Known limitations

• Worker computes no score yet; Upwork client field names “marked VERIFY pending first-run introspection.”
• Phase-1 persistence is localStorage (no shared multi-user store).
• Web deploy must run on Linux (OpenNext @ast-grep/napi fails on Windows) — use Cloudflare Workers Builds.
• Extensive Upwork OAuth gotchas documented (CSRF/state, scope-freeze, redirect byte-match).

Future opportunities

Wire live Upwork scoring; AI-assisted qualification; consume the Website Assessment output as evidence for website opportunities; proposal assistant + outcome learning.

Relationship to the wider AI venture ecosystem

Layer-4 application serving Inexis Digital + Inexis Consulting — the inbound counterpart to the outbound outreachagent. It consumes Intelligence Products (pinned submodule) and is designed to consume the Website Assessment output. See the Applications & Agents system.

Links

• Source: github.com/azwaanameer/leadplatform (private)
• System: Applications & Agents
• Registry row: repo registry