Solutions Architect for growth-stage startups.Now shipping three production AI systems with Claude.
I’m Shirley. Pre-sales SA and CSE leadership at Demandbase and Clearbit (acq. HubSpot)— tier-one customers Stripe, Slack, Meta, Intercom, Segment. Today, in CDMX: five live businesses and AI-native systems I ship with Claude daily.
I’m not actively searching.
The honest version
On the obvious question first — yes, I have been busy building and supporting five businesses in Mexico City. The reason I can step into a full-time role now is that the last two years of work have been about exactly that: building the org, the processes, and the platform (RIOS) so the businesses run without me as a single point of failure. My partner co-runs them with a fully capable team behind her; the day-to-day continues whether I’m involved or not. Stepping back from operations was the goal. Anthropic was attractive to me a year ago too — the timing just wasn’t right then. It’s starting to feel right now.
What pulled me toward Anthropic is twofold. This role in particular reads like an unusual fit with the work I’ve actually done — pre-sales Solutions Architecture and post-sales engineering ownership for growth-stage tech customers, founder-led technical conversations, and the architectural advisory work that lives between technical evaluation and production deployment. And Anthropic itself is, by almost any measure, one of the most attractive places in tech right now — the work, the team, and the trajectory all line up. I’ve been using Claude as a daily operator across my businesses, and the rate of capability gains release-over-release has been hard to look away from — it’s rare to watch a product redefine what’s possible this quickly. The mission — advancing safe and beneficial AI systems — is one of the few in tech I find genuinely worth spending the next chapter of my career on. I want to be at Anthropic. This role is the cleanest fit; if the team thinks I’d land harder somewhere adjacent, I’d hear that conversation too.
After a decade of doing this, my honest evaluation criteria for any job have stayed the same three things, in this order: a place with an overwhelming amount of opportunities to continue learning, a smart team, and cool products. That framework has never led me wrong. Anthropic clears all three.
Solutions Architect for a decade at Demandbase and Clearbit (Series A through HubSpot acquisition) — exactly the kind of growth-stage tech companies the JD describes as customer audience. At Clearbit I was Head of Solutions Architecture and Customer Success Engineering, owning both pre-sales and post-sales technical work end-to-end. That ownership forced a level of detail in the sales cycle pure pre-sales teams don’t carry — every scoping commitment had to be one my own delivery team could honor. Recently I’ve shipped three production AI systems with Claude — RIOS, CasaRuta, and CapitalRuta — building the same architectural patterns Anthropic’s startup customers face. I also run five food and beverage businesses in CDMX with my partner; the operator chair is what taught me what production reality actually demands of a system. That’s the combination I’d bring on day one — full-lifecycle ownership at growth-stage tech, plus practitioner-level Claude experience.
I’m applying for one specific role — here’s the JD,line by line, against my experience.
The four responsibility lines from the JD, mapped to the closest evidence in my actual experience.
- JD line 1
Trusted technical advisor through the Claude adoption journey — partner with Account Executives, translate startup requirements into technical solutions, support evaluation through deployment and expansion
My evidenceEmbedded technical advisor has been the job since 2014. At Demandbase as Solutions Engineering Lead, at Clearbit running both the SA and CSE org for tier-one accounts (Stripe, Slack, Meta, Intercom, Unity, Segment in their growth-stage years). Owning both pre-sales and post-sales technical work meant the sales conversations had to scope solutions my own team could actually deliver — no overclaiming, no half-baked architecture promises in the scoping call. AE partnership was daily work: reps trusted me in the room because what I committed to in pre-sales shipped without surprises post-close.
- JD line 2
Win technical evaluations — help startups develop evaluation frameworks to measure Claude's performance for their specific use cases
My evidenceEval design for AI products is where I’ve spent the last two years. RIOS, CasaRuta, and CapitalRuta each required me to build the eval framework from scratch — golden datasets, regression suites, drift detection, and the human-review boundaries that decide where the system can and can’t be trusted. Same pattern carried over from a decade of enterprise technical evaluations: define what ‘working’ means in the customer’s context, instrument for it, and make the technical case empirically rather than rhetorically.
- JD line 3
Build technical credibility with founders, founding engineers, and startup engineering teams — speak their language, understand build patterns, guide architecture decisions
My evidenceI’ve been a founder-shaped operator for the last two years and a technical leader at growth-stage startups for a decade before that — Vero (Series A), Demandbase (mid-stage), Clearbit (Series A through HubSpot acquisition). The constraints, velocity, and trade-offs early-stage teams make are the constraints I live in daily. And I’ve made every architecture decision a Claude-native company faces: deterministic vs agentic boundaries, eval discipline, when to invest in observability, when human review is non-negotiable, where the rules end and the model begins.
- JD line 4
Hands-on building and deploying LLM-powered applications in production — context engineering, evaluation frameworks, modern AI architectures
My evidenceRIOS, CasaRuta, and CapitalRuta are three production LLM-powered systems I designed, built, and operate — across measurement infrastructure (RIOS), product comparison and eligibility logic for 104 mortgage products (CasaRuta), and SMB capital matching (CapitalRuta). Context engineering, eval-loop design, retrieval boundaries, prompt versioning, and the deterministic/agentic split are decisions I’ve made deliberately for each. Comfortable with Python and the Anthropic API/SDK as my daily build surface. Not theoretical — running today.
- JD line 5
Track record selling technical products in competitive markets + strong technical communication for founders and engineering teams
My evidenceSelling technical solutions has been the job — closing enterprise deals at Demandbase against Marketo, Bizible, and the rest of the ABM market; defending Clearbit’s data layer against ZoomInfo and Apollo in tier-one bake-offs. The honest caveat: I haven’t carried a quota formally. I’ve been the technical lead on more competitive evaluations than I can count, and I’ve earned trust with founders and engineering teams the way the JD describes — by speaking the language, knowing the build patterns, and not bluffing on what I don’t know. Quota-bearing OTE is the one piece I’d be ramping into; everything else is daily work.
Three builds in production.
Operational intelligence across a multi-vertical portfolio of operating businesses
An executive dashboard and data-completeness layer that unifies POS, inventory, and accounting signal across five operating businesses.
Each business in the portfolio runs on its own combination of POS, inventory, accounting, and staffing tools. Reports disagreed by 5–15% on any given day. The decisions that mattered — staffing tomorrow, ordering next week, whether last week’s pricing change worked — were being made on lagged, incomplete, or inconsistent data.
- 01Ingestion
Vendor adapters (POS and PDF/CSV imports) backfill historical windows on a rolling cursor and watch for late-arriving rows. Every ingestion run is logged and safe to re-run — running the same job twice produces the same result, never duplicate data.
- 02Completeness
A first-class completeness layer per entity (empresa / marca / ubicación) — every dashboard view declares what data it depends on and refuses to render misleading numbers when something is missing or stale.
- 03Reconciliation
Same-grain comparisons across POS, accounting, and bank deposits. Drift is surfaced as an exception rather than averaged into a misleadingly clean total.
- 04Executive layer
A small, opinionated set of executive views — daily revenue and trend, ticket size, item mix, labor against forecast, supplier variance — built around what an operator actually decides on, not what's easy to chart.
- 05Agent layer
Claude on top, scoped to an explicit memory of business structure, vendors, and known anomalies — used for explanation, summarization, and drafting weekly operating notes that are reviewed before they leave the system.
Used. Claude reads the warehouse for natural-language queries, anomaly explanation (“why is Tuesday low?”), and structured drafting of the weekly operating note. Every Claude touchpoint sits behind a deterministic data fetch and a human review step before the output is acted on.
- Month-end close happens in real time instead of after a three-week wait — books are continuous, not retrospective.
- We cut waste 5% in the first month, driven by real-time margin and cost alerting.
- AI extracts invoices (CFDIs, PDFs, XMLs) and matches them to bank deposits and POS sales — manual reconciliation is gone. For us, that’s two full-time jobs eliminated.
- Payroll and labor-cost-as-percent-of-revenue tracked continuously per location, not at the end of the period.
- Multi-entity, multi-brand, multi-location P&L in one engine — plug a new business in and the executive layer is online in days.
Ten years building, integrating, and deploying the tools sales, marketing, and revenue teams rely on.
Click any row to expand
So, what’s the easiest way to become a founder?
- Move to a country where you barely speak the language.
- Take on a tax code that takes years to learn — and rewrites itself while you do.
- Navigate an employee system you think you know — but in time realize you have no clue about.
- Make sure it’s not tech.
After a decade in tech, I’d done most of what I’d set out to do — except build a business from zero, on my own terms. So my partner and I moved to CDMX and together started a small portfolio of operating businesses. The systems thinking carried over. Only the inputs changed — customers, staff, regulators, supply chain.
Five live locations across CDMX, all operating today. The startup skillsets I built in tech — instrumentation, measurement, systems thinking, ruthless prioritization — are the direct reason we’ve been able to scale from one location to five, and what sets us apart from most operators in the category.
- 01Early on — work at a big company.
- 02Build and lead great teams.
- 03Make meaningful contributions to a company that lead to a successful exit event.
- 04Start my own thing and make it successful.
- Attribution and measurement
- Targeting
- Marketing
- Negotiations
- Sales
- AI-native systems design
- Multi-entity ops architecture
- Planning and Projections
- Prioritization
- Team building
- Operational excellence
- Builder-operator translation
Now: Applied AI Architect,Anthropic.
If the JD lands — let’s talk
Ten years architecting for growth-stage tech. A chapter operating from the customer chair — in a country and tax code I had to learn from scratch. Recent work shipping production AI systems with Claude. This is the role that brings the three together — helping the next wave of Claude-native startups go from evaluation to scale.
The resume’s below. The cover letter is this page. The rest is whatever you need to ask me.