role
Seth — Lead Architect JD
Seth Shoultes's Lead Architect role definition: end-to-end technical ownership of the brain, data pipeline, AI/ML architecture, engineering team, and vendor decisions for MemberIntel.
Role: Lead Architect, MemberIntel
Incumbent: Seth Shoultes
Reports to: Blair Williams, CEO of MemberIntel
Peers (also report to Blair): Cindy Thoennessen (Product Lead, MemberIntel) · Santiago Perez Asis (Project Manager, Cross-Caseproof)
Effective: May 2026
References: MemberIntel SPEC v1 (/SPEC-v1.md)
Mission
Own the technical product for MemberIntel end-to-end. Architecture, the data pipeline, AI/ML model design, the brain (global + per-customer), the analytics engine, the API, security, and performance. Direct the engineering team. Be the technical voice that customers, partners, and the Caseproof team trust on AI and data infrastructure decisions.
You’re the most advanced AI architect at Caseproof. The role is built to maximize that strategic technical contribution while pairing you with the structural support to ship clean, on-time, on-quality work.
Authority structure
Blair (CEO, MemberIntel) holds:
- Product strategy, target customer, 18-month roadmap
- Final say on PRDs (you provide technical feasibility input)
- Final say on product design (you provide technical constraints input)
- Final say on architecture decisions for material choices (you propose)
- Pricing strategy
Lead Architect (Seth) holds:
- Technical architecture and implementation
- Engineering team direction and code quality
- Sprint scope and engineering velocity
- Engineering hiring decisions
- Vendor / tooling decisions
Cindy (Product Lead) holds:
- Cross-functional execution, marketing site, content, beta program, privacy/compliance, customer onboarding
- Quality bar / “are we shipping or not” gate
Santiago (Project Manager) holds:
- Sprint cadence and ticket tracking infrastructure
- Cross-rock dependency tracking
What you own
1. Technical architecture and design (per SPEC §6, §8, §12)
- The brain — both tiers:
- Global brain: chunking, embedding, indexing, retrieval, content authoring system, version control
- Per-customer brain: storage, partitioning, retrieval, update mechanisms, version history
- Cross-pollination job (weekly/monthly): Claude-driven candidate drafting + content-lead review queue
- Data pipeline architecture:
- Customer’s MemberPress site sync (live for Pro, monthly snapshot for Free)
- Stripe Connect / Stripe OAuth integration
- Public site analysis pipeline (Claude-based, weekly cached, on-demand for Pro)
- Per-customer warehouse (Postgres) — schema, partitioning, isolation
- AI/ML architecture:
- Model routing (Sonnet for Pro, Haiku for Free) — server-side enforced, never user-configurable
- Tool surface implementation (
query_customer_metrics,analyze_site,search_global_brain,search_customer_brain,update_customer_brain) - Prompt versioning and the AI eval suite
- Hallucination guards (citation requirements, data-ID grounding)
- Continuous improvement loop from thumbs feedback (no fine-tuning in V1)
- Entitlement layer — server-side service that owns tier model: who can do what, token budgets, sync cadence, model routing. Single source of truth.
- Quota tracking — per-user, per-month counters; visible to user and support staff.
- Cost controls (per SPEC §5.5):
- Hard token budget caps server-side (Free: tight; Pro: generous-but-not-unlimited)
- Aggressive caching (site analysis 7-day, insight cards until refresh, digest pre-computed)
- Brain retrieval pre-filtered before LLM (no full-corpus dumps)
- Abuse prevention (rate limits, anomaly flagging)
- Security architecture:
- Per-tenant data isolation in the warehouse
- GDPR / CCPA compliance from day one (data export, deletion pathway)
- Auth (OAuth via MP license + standard email/password fallback with MFA option)
- Performance and scalability:
- Sync pipeline resilience (rate limits, schema changes, customer plugin updates)
- Vector store choice (pgvector to start; Pinecone if scale demands)
- Hosting decision (GCP vs Heroku — your call per SPEC Open Q8)
- Vendor / tooling:
- Anthropic API directly (no LangChain in V1)
- Stripe integration coordination with Ally Roger
- Vector store choice (per SPEC Open Q9)
- Telemetry / observability stack (per SPEC Open Q11)
2. Engineering leadership
- Direct reports:
- Ronald Reymundo (WP Dev) — interim code quality enforcer
- Senior AI Engineer (new hire — your search, target close mid-June)
- Matrixed engineering support:
- Mahmoud Saeed (in IPJ under Thomas) — ad hoc data layer / senior engineering consultation if needed
- Thomas Levy (IPJ Lead) — data layer engineering consultation
- Hiring:
- Senior AI Engineer search and final call (Danielle handles cultural / operational fit interviews + onboarding logistics; you decide engineering fit)
- Future engineering hires as MemberIntel scales
- Sprint planning — collaboration with Santiago on cadence; you decide scope and velocity
- Code quality — interim through Ronald; long-term through Senior AI Engineer once ramped
- Pairing, mentoring, career development for direct reports
3. Build and ship (per SPEC §13 phases)
Phase 0 (Weeks 0–4): architecture decisions, vendor selection, hosting choice, Senior AI Engineer hiring kickoff.
Phase 1 (Months 1–3): Operator core build:
- Auth, customer-data ingestion (MP + Stripe), per-customer warehouse + schema
- Global brain seed + indexing
- Per-customer brain scaffolding
- AI tool surface
- Public site analysis pipeline
- Chat advisor with citations and feedback capture
- Tier-gated model routing (Haiku/Sonnet)
- Entitlement service + quota tracking
- Standard dashboard with tier-gated insight cards
- Free vs Pro UI differentiation
- Self-serve billing integration with Ally
- Privacy controls + GDPR/CCPA data export + deletion
- i18n scaffolding (English + Spanish + German)
- Milestone (Month 3): internal beta with 10–20 hand-picked MP customers across both tiers
Phase 2 (Months 4–6): Polish + launch:
- Cross-pollination job (weekly schedule + content-lead review queue)
- Weekly digest email generation (Haiku/Sonnet)
- AI eval suite + prompt regression tests
- Cost monitoring + per-customer token caps
- Free-tier signup flow polish
- Upgrade-funnel telemetry
- Cost-monitoring dashboards
- Public launch
4. Cross-functional technical coordination
| Partner | What you coordinate |
|---|---|
| Cindy Thoennessen (Product Lead) | PRD feasibility input, bug triage prioritization, technical input on product decisions |
| Santiago Perez Asis (PM) | Sprint cadence, dependency tracking |
| Paul Carter (MemberCore Lead) | Data access agreements; MP-side API surface required for V1 sync (per SPEC Open Q7) |
| Thomas Levy (IPJ) | Data layer engineering consultation; Mahmoud as ad hoc resource if needed |
| Russ Williams (Design Lead) | Meo’s home manager — design coordination for MemberIntel UI |
| Ally Roger (Payment Service) | Stripe billing integration, plan upgrades/downgrades/dunning |
| Danielle Lea-Jones (Operations) | Senior AI Engineer hire — cultural/operational fit interviews + onboarding logistics |
| Outside privacy counsel | Data architecture compliance (engaged by Cindy; you provide technical input) |
5. Strategic technical voice
- Surface architectural risks and opportunities to Blair
- Stay current on AI/ML state of the art (Anthropic releases, alternative model providers, retrieval techniques)
- Vendor relationships with Anthropic, data infrastructure providers
- Technical content / thought leadership (Cindy will pull on this for marketing — you provide; Cindy directs the campaign)
6. Reporting
- Weekly technical status to Blair
- Architectural decision records (lightweight, but written) for material choices
- Sprint review participation
- L10 attendance
What you do NOT own
- Product strategy, target customer, 18-month roadmap — Blair
- Final approval on PRDs — Blair (you provide feasibility input)
- Final approval on product design — Blair (you provide technical constraints)
- Pricing strategy and packaging — Blair (you implement)
- Marketing positioning, GTM, content — Cindy + Curt
- Customer onboarding flow design — Cindy
- Privacy / compliance program — Cindy (with outside counsel)
- Project management, sprint tracking infrastructure — Santiago
- Cross-functional coordination outside engineering — Cindy
- Quality bar / ship gate — Cindy holds the gate (you provide technical readiness; she makes the call)
- MM/AA/WL legacy work — handed off
Structural support (the team is designed around your strengths)
You are the most advanced AI architect at Caseproof. The team structure is intentionally designed so you can focus on architecture, AI/ML, and engineering leadership without being pulled into adjacent work. This is a feature.
- You don’t track tickets — Santiago does that. Use freed time on architecture.
- You don’t run code review yourself early on — Ronald enforces baseline quality; the Senior AI Engineer takes the senior review role once ramped. If quality slips, Mahmoud is available ad hoc through Thomas.
- You don’t coordinate marketing or content — Cindy does. Brief her on what’s true; let her translate to copy.
- You don’t negotiate cross-functional priorities — Cindy does, with backstop from Blair.
- You don’t make the ship/no-ship call — Cindy holds the gate, with your technical-readiness input.
The structure is built to maximize your strategic technical output. Use it.
Critical role norms
- Architectural decisions get written down. Lightweight ADRs (Architectural Decision Records) for material choices — vendor selection, model routing, brain schema, entitlement layer design. Future-you and the team need these.
- Citations are non-negotiable. Per SPEC §8.4, every AI response must cite the data it references. Hallucinations on financial data are trust-killers (per SPEC Risk #7). Build the citation discipline into the prompt and eval suite.
- Free-tier server-side enforcement of model routing. Never accidentally route a Free user to Sonnet. Single source of truth, enforced at the entitlement layer.
- Cost discipline first, scale second. Per SPEC §5.5 — hard token caps, aggressive caching, weekly cost-per-cohort review during early launch. Don’t optimize for performance at the expense of unit economics.
- Disagreements with Cindy go to Blair within 48 hours. No silent escalation drift. If something not in the decision-rights matrix comes up, surface it fast.
- The Senior AI Engineer hire is the most consequential decision of Q2. Treat it accordingly — invest the time. Bias toward someone who can be a senior code reviewer + architect counterpart, not just a strong individual contributor.
- No fine-tuning in V1. All learning happens through brain content + retrieval. (Per SPEC §3 non-goals and §8.4.)
Success measures (12-month)
| Measure | Target |
|---|---|
| MemberIntel V1 GA shipped | On or before mid-October 2026 |
| Internal beta milestone | Month 3 (per SPEC §13) — 10–20 customers using the system |
| Senior AI Engineer hired and ramped | Mid-June close, productive by August |
| AI eval suite operational | Before launch — proving MP-specific advantage over baseline LLM (per SPEC Risk #1 mitigation) |
| Cost-per-Free-user (steady state) | At or below $1.10/mo (per SPEC §5.4 estimate) |
| Cost-per-Pro-user (steady state) | At or below $12/mo (per SPEC §5.4 estimate) |
| System reliability at launch | 99.5%+ uptime in the first 30 days post-GA |
| Hallucination rate on financial-data answers | < 1% on the eval suite |
| Architectural decision records | One ADR per material choice, all in the repo |
Reporting cadence
| Cadence | Audience | Format |
|---|---|---|
| Weekly | Blair | Technical status (5–10 bullet points) |
| Bi-weekly | Cindy + Santiago | Sync meeting (30 min) |
| Monthly | Leadership team | L10 attendance + rock report |
| Quarterly | Blair | Architectural review (90-min deep-dive) |
What “good” looks like in this role
- The architecture you choose at the start is still serving the product 18 months in
- Engineering velocity is steady — not heroic sprints followed by burnout
- The brain (global + per-customer) is the moat the SPEC promises — not a thin wrapper around generic LLM responses
- Cost-per-user stays at or below modeled targets through 50K+ free users
- The Senior AI Engineer hire is a strong peer who you trust with senior reviews
- Customers feel the AI advisor knows their site (citations, grounding, per-customer brain working as designed)
- You have time to think — the structural support around you (Cindy, Santiago, Mahmoud-as-backstop) is doing its job
- The team enjoys working with you — your easygoing style is a force multiplier when paired with Cindy’s delivery discipline
Document version: Draft v1 — to be reviewed with Seth before finalization. Decision-rights matrix requires sign-off from Blair, Cindy, and Seth before this JD goes operational.