Sebastián Arduz
EN ES

Juan Sebastián Arduz Espada

Full-Stack Engineer (TypeScript) · Automation, integrations and outbound systems

Sucre, Bolivia (UTC-4) · Remote, US Eastern or EU hours · Available now · Spanish native, English C1

Download CV (EN) CV en español Email GitHub LinkedIn
Portrait of Juan Sebastián Arduz Espada
80
PRs merged in 114 commits to a production TypeScript monorepo, Jul to Oct 2026
13
scheduled monitors with Slack and email alerts
9
Microsoft 365 tenants built, about 225 mailboxes with SPF, DKIM and DMARC
22,728
geo-clean contacts in 29 lead builds for 32 clients
83
PRs merged on a hiring funnel and training portal (second contract)
378,605
emails across 302 campaigns in the 90-day review that found the angle that worked

Every number comes from git history or an internal report. The code lives in private company repositories, so this page shows the systems and the results, not the source.

Summary

Full-stack TypeScript engineer. For one year I was the in-house engineer and automation owner of a US B2B outbound agency, working alongside its technical founder: 80 pull requests merged to a production Next.js, Prisma, PostgreSQL and Trigger.dev monorepo, 13 scheduled monitors, root-cause fixes shipped to production, and the GoHighLevel, Instantly and Microsoft 365 automation layer around it. Daily Claude Code user; nothing merges without guard checks, an independent review pass and my own read of the diff. Java and Spring Boot before that.

Experience

Full-Stack Engineer (TypeScript) · Leveraged Operations (Local GTM), US, remote contract

December 2025 to October 2026

In-house engineer and automation owner, working alongside the technical founder, for an outbound sales operation (cold email on Instantly, SDR cold calling on GoHighLevel, lead generation) serving US local-service businesses. Dec 2025 to Jun 2026: client integrations and data work, Microsoft 365 email infrastructure and lead pipelines. Jul to Oct 2026: 80 merged PRs to the core platform, 114 commits in all.

  • 80 PRs merged to the production monorepo in 114 commits: Next.js 16 (App Router) and React 19, oRPC API, Prisma on PostgreSQL (Supabase), Trigger.dev jobs; 19 SQL migrations and 16 new guard-check files in a codebase of 306 oRPC procedures and 227 Prisma models. Each PR passed Biome, type-check, build and an independent AI review before merge; each deploy confirmed against the live build SHA.
  • Rebuilt the team chat into a Slack-style chat in 16 PRs over 5 days: DMs, mentions, reactions, quoted replies, image paste, in-app and OS notifications, WCAG-AA avatar palette, later presence and manager email alerts.
  • Upgraded the SDR lead-approval workflow in 8 PRs merged in one day (3,843 lines, 74 of 74 guard checks): AI first drafts behind a placeholder guard, reject-reason presets, per-rep 7-day stats, automatic 3-day callback tasks.
  • Wrote 13 scheduled Trigger.dev monitors: bounce guardrail (alerts above 3% on 200+ sends in 7 days, informs and never auto-pauses), reply-rate decay, lead runway, live campaigns with zero leads, paying clients with zero fulfillment, nightly dial readiness.
  • Root-caused production incidents and shipped tested fixes: dialer queues leaking other clients' leads (10 leaks confirmed by a per-rep audit), a nightly job that aborted every night but reported COMPLETED for four weeks, 127 renewed domains falsely flagged as expired (now cross-checked over RDAP), and a caller-ID warning backtested against 6,882 real call evaluations with 0 mismatches.
  • Provisioned the cold-email infrastructure: 9 Microsoft 365 tenants (about 225 shared mailboxes) with SPF, DKIM and DMARC through the Cloudflare API, 26 PowerShell scripts and 2 Playwright bots (Instantly OAuth connection at 97% first-try success); audited 497 domains and 2,646 IPs against DNS blocklists; purged 46,192 non-responders across 42 campaigns by API with a backup first, keeping all 1,538 repliers.
  • Built the lead pipeline from scrape to draft campaign: D7 and Apify scraping with cost caps, normalization and dedupe, a verification waterfall (MillionVerifier, then a Reacher SMTP pass added after proving the single pass let 57 of 64 real bounces through, then Reoon), LLM scope screening, Instantly draft campaigns by API; 22,728 geo-clean contacts in 29 builds for 32 clients.
  • Campaign and reply tooling: a copy linter that renders spintax as the prospect sees it (first sweep flagged 71 drafts with 498 failures and exposed an Instantly API bug), an admin escalation board for cold-email replies, and a 90-day review (302 campaigns, 378,605 emails, 4,063 replies, with an LLM classifying a 680-reply sample) showing the referral angle produced 39 of 48 opportunities.
  • Hardened public and paid endpoints: per-organization daily quotas and a global circuit breaker on paid fan-outs, rate limits and input caps, security headers, rate-limited passcode login.
  • Technical support for the SDR team: GoHighLevel dialer provisioning with least-privilege scopes, local numbers and custom fields for each cohort, designated approver of the lead and email queues, 10 runbooks and SOPs.

Full-Stack Engineer (Node.js) · Chatter Champions, creator-talent agency, remote contract

May to August 2026, in parallel with the role above

Hiring funnel and training portal for a remote chat-support workforce: Express, PostgreSQL (Supabase) with Google Sheets dual-write, Airtable, GoHighLevel, Trigger.dev, xAI Grok, Resend, Telegram, Dokploy.

  • 83 merged PRs (100 commits on main, +11.3k lines) across product features, automations and admin tooling under a protected-main PR workflow.
  • Fixed the incident behind a 5-of-8 interview no-show day: reminder emails carried broken Google Meet links. Created the missing custom field via API, backfilled 28 bookings and shipped the Trigger.dev fix (meeting-link sync plus a one-booking-per-contact rule).
  • Built the AI interview simulator and the interview-integrity layer: capture-only chat with a cost cap, transcript review, drop-off and abandonment recovery jobs, no-show detection with a daily digest, auto-ban of rejected applicants, an intake screen-out rolled out as an A/B test.
  • Admin and HR tooling: manager role with gated writes, hired-by-contract-date view, trainee-completion view, daily HR alerts, Airtable sync, shift template editor.

Freelance Software Developer (Java, Spring Boot) · Sucre, Bolivia

November 2024 to November 2025

  • Car dealership management system (Feb to Apr 2025): JavaFX desktop app on Spring Boot (Spring Data JPA, Spring Security) and MySQL, with full CRUD for customers, vehicles, quotes, maintenance records and reminders. Source on GitHub.
  • Private clinic records system (Nov 2024 to Jan 2025): JavaFX desktop app on Spring Boot and MySQL for patient records, clinical history, vital signs and search. Source on GitHub.

Case studies six systems, with the numbers behind them

Scheduled monitors and alerting on Trigger.dev

Thirteen scheduled jobs that catch silent failures: campaigns running out of leads, bounce spikes, missed domain renewals, jobs that abort but report success. Two severity tiers, 24-hour de-duplication, guard checks with control cases.

Details
  • Domain renewal sync cross-checked over RDAP: a shadow run took the "expired" count from 128 to 1 (127 domains carrying 281 active mailboxes).
  • A nightly job had aborted every night for four weeks while reading COMPLETED; aborts now fail visibly without a retry storm.
  • Caller-ID rule backtested against 6,882 real call evaluations with 0 mismatches before going live.
  • Lesson: alert on rates, not counts, and make "clear" distinguishable from "broken".
TypeScriptTrigger.devPrismaPostgreSQLSlack

A Slack-style team chat for an SDR academy

Sixteen merged PRs in five days: direct messages, mentions, reactions, quoted replies, image paste, notifications and presence inside an internal Next.js portal, with the access rule enforced on every call and guarded by a check file.

Details
  • Rooms derived from the user's profile, no membership table; DMs keyed by the two sorted user ids so both sides open the same room without a migration.
  • Notifications with zero extra requests: the portal already polled unread state, so the tab title, a toast, an OS notification and a chime hang off that poll.
  • Presence shipped behind a kill switch that is on by default, because production environment variables were out of reach.
Next.jsReactoRPCTanStack QueryR2

Lead pipeline and email deliverability

From sourcing to a draft campaign: Google-Maps-based search and Apify with cost caps, normalization and dedupe, a measured verification waterfall, an AI scope filter, Instantly drafts by API and a copy linter. Plus the Microsoft 365 infrastructure the emails go out from.

Details
  • A single verifier pass let 57 of 64 real bounces through, so a second SMTP pass was added (it catches about 17% of bounces) and a third tool tested against labeled ground truth, 25 of 25.
  • 22,728 geo-clean contacts in 29 builds for 32 clients, 27 of 29 passing a 40-row spot-check with controls.
  • Nine Microsoft 365 tenants from zero, about 225 shared mailboxes, DNS written through the Cloudflare API, 26 PowerShell scripts, 2 Playwright bots, OAuth connection at 97% first-try success.
  • DNSBL audit of 497 domains and 2,646 relay IPs: the listings sat on domains, not on sending IPs. Placement study over about 1.5M emails: 47.4% spam when SPF or DMARC failed versus 3.2% when they passed.
  • Self-hosted contact database: PostgreSQL 16 behind PostgREST and Nginx with TLS, one query tuned from 10.3 s to 0.26 s.
Node.jsPowerShellPlaywrightInstantly APICloudflare APIMicrosoft 365

Dialer operations on GoHighLevel

The engineering behind a commission-based SDR team's power dialer: queue loads per shift, a nightly scheduler, caller-ID rules, provisioning and real-time support.

Details
  • Queues leaking other clients' leads: a 36-agent adversarial audit confirmed 10 distinct leaks; the fix shipped the same day it was measured, after three review rounds.
  • Caller ID: 93% of calls were going out on the shared default line; after correcting 24 briefs and rebuilding the warning for Local Presence Dialing, default-line use fell to 0 to 4%.
  • Scheduler guards: never auto-dial a client without a number or a brief, respect booked time off, list every skip in the nightly email.
  • Provisioning: local numbers bought and assigned through the API, least-privilege users for each cohort, an offboarding script that revokes access across six surfaces and is dry-run by default.
GoHighLevel APITrigger.devPrismaNext.js

How I work with Claude Code

AI makes you faster; tests, reviews and honest numbers make you right. Guard checks with control cases, adversarial multi-agent reviews before merge, type-check plus Biome plus build, and a deploy confirmed against the live build SHA.

Details
  • A 15-agent review of a lead-magnet calculator found 6 real bugs in the first commit, all fixed before merge.
  • Authored skills with their own regression suites: a cold-email copy reviewer with the company's 13 hard rules and a 9-case harness graded against 365 real campaigns; a positive-reply triage skill. 62 ops scripts that load their own keys, 10 role runbooks.
  • A monorepo index (306 oRPC procedures, 227 Prisma models) that cut the agent bootstrap file from 921 to 463 lines, and a code graph measured in a 16-run A/B: it helps "who calls this" lookups and hurts end-to-end traces, so it is used only where it wins.
  • Multi-agent workflows where one pass is not enough: 10 scoping agents plus 2 auditors, 24 judges for copy QA, 36 agents for the dialer audit.
Claude CodeAnthropic APILLM-as-judgeCI guard checks

Hiring funnel and training portal (Chatter Champions)

Application, test, interview, offer and contract, then a six-day training portal with an AI practice chat, for a remote chat-support workforce. 83 merged PRs in under three months, in parallel with my main role.

Details
  • Fast-track interview simulator: capture-only AI chat with a per-run cost cap, transcript review with a hire or not-a-fit decision, drop-off recovery and a five-day abandonment job. Twenty PRs in two days once the design was agreed.
  • Interview integrity: serial-rescheduler tracking, automatic no-show detection with a daily digest, auto-ban of rejected or no-show applicants, an intake screen-out rolled out as an A/B from 30% to 100% of traffic.
  • The incident: five of eight interviews no-showed because reminder emails carried a broken meeting link. Confirmed the data, created the missing custom field through the API, backfilled 28 bookings, shipped the fix, and corrected my own first status message with the client the same day.
Node.jsExpressPostgreSQLTrigger.devGoHighLevelxAI Grok

AI workflow, in one paragraph

I write the spec first, let agents attack it for scope and data risks, then implement in independent slices with tests while I review diffs instead of typing. Schema, state machines and external integrations stay under my own hands; agents own the plumbing, UI states and tests. Everything merges behind CI guard checks and a judge panel with different lenses, and anything the agents could not test gets a monitor instead of hope. Every line that reaches production has been read by me.

Skills

Languages
TypeScript / JavaScript (Node.js), SQL, Java, Python, PowerShell, Bash
Backend and data
Next.js 16 (App Router), REST APIs, oRPC (on Hono), Express, Prisma, PostgreSQL (Supabase, PostgREST), MySQL, Spring Boot, background jobs and scheduled tasks (Trigger.dev), webhooks, OAuth integrations, data pipelines (ETL, dedupe, verification)
Frontend
React, Tailwind CSS, shadcn/ui, responsive UI, WCAG-AA color contrast
Outbound and GTM systems
Cold-email infrastructure (Microsoft 365 / Exchange Online tenants, mailbox provisioning, warm-up, SPF/DKIM/DMARC, DNSBL audits, inbox placement analysis), Instantly API v2, email verification waterfall (MillionVerifier, Reacher, Reoon), list building and ICP scoping (D7, Apify, Clay, LLM scope filter), reply classification and routing, copy QA and spintax linting, GoHighLevel CRM and power dialer (user provisioning, workflow design and audit, custom fields, caller ID drift detection under Local Presence Dialing)
AI engineering
Claude Code (skills, hooks, multi-agent workflows, scheduled routines), Anthropic and OpenAI APIs, xAI Grok, prompt engineering, LLM-as-judge evaluations
Engineering practice and infra
Git / GitHub PR workflow, code review, CI/CD (GitHub Actions), automated guard checks, Biome, pnpm monorepos, Playwright, Dokploy (Docker-based deploys), Nginx, Linux VPS (Hetzner), Cloudflare API, Airtable, Google Sheets / Drive APIs, Resend, Slack, Telegram, production monitoring and alerting, incident response, technical documentation (SOPs, runbooks)

Education and certifications

Contact

Remote roles and contracts, US Eastern or EU hours, available immediately. The fastest way to reach me is email.

jsebastianarduzespa@gmail.com LinkedIn GitHub CV (EN, PDF) CV (ES, PDF)