fireharp.com/cv

Staff/Principal product systems engineer

I turn prototypes and messy business workflows into production systems: billing, payments, funnels, product analytics, operational tooling, AI workflows, observability, eval harnesses, and CI quality gates. Remote from Spain/EU; strongest fit is senior IC or small-team builder work with meaningful equity.

Python · TypeScript · React · Node.js · Go · Django/FastAPI · PostgreSQL · BigQuery · GCP/AWS · Cloudflare Workers · Playwright · LLM workflows · MCP · agentic tooling

// select projects

A few relevant systems and project surfaces. The links are public-safe entry points; private/raw evidence stays local.

G-Plans / FuturHealth

First hire to technology owner across checkout, billing, payments, funnels, product analytics, internal tools, and production support.

PixelFixer

Ad-pixel observability, landing-page perception extraction, synthetic traffic calibration, and inspectable event ledgers.

Figma Harness

Agent-generated UI code brought under Playwright screenshots, visual QA, fix queues, and review gates.

Tinkershop / evidence workflows

Local project evidence, policy gates, generated dashboards, package assembly, and repeatable agent-assisted quality checks.

Knify / canvases

Agentic workload surfaces with React/Vite, Cloudflare Workers, SSE job logs, webhook routing, and reviewable output.

Mumbli / agent workflow

macOS speech-to-text and voice-to-agent workflow for turning spoken specs into bounded engineering tasks.

// timeline

1999
2016
2025
now
06

Hands-on product systems

G-Plans -> current
05

Agentic tooling

LLM systems / semantic analysis / agents
04

Scale-up transition

FuturHealth
03

Bootstrapped systems

G-Plans / Moon
02

Operating reality

southern China / suppliers
01

Foundations

programming / physics / web

recent zoom band

2025 - now

// operating style

How I tend to work

A few stable patterns that show up across product, platform, data, and agentic systems.

01 / own the loop

Define the system, ship, instrument, read the evidence, and tighten the next pass.

02 / small teams

Deliver under real budget/headcount constraints without hiding behind platform theater.

03 / economics

Build around payback, retention, checkout risk, pricing, billing, and operational reality.

04 / hands-on

Current focus: agents, LLM workflows, harnesses, voice input, and review surfaces.

// FuturHealth: hands-on platform transition

Recent developer bridge mission

Inherited-stack scale, new-stack migration, production continuity, and handover as a hands-on developer/technical lead before moving into agentic tooling and product systems.

Jan 2025 - Apr 2026
FuturHealth

Senior Software Engineer, Backend and Production Systems

  • Hands-on IC/transition engineer after the GP team rolled off and the US team ramped; covered legacy fixes, support/debugging/triage, API review, runbooks, and AI-assisted tooling.
  • Maintained inherited production systems while payments, funnels, support workflows, and user flows moved across stacks; shaped Go/OpenAPI/Temporal/PDF/DataDog work from production feedback.
  • Worked around PostHog flags/session data, Braze/Zendesk/support context, CRM, payment, provider, alert, and lifecycle systems where measurement quality affected operations.
Dec 2023 - Dec 2024
FuturHealth

Principal Software Engineer / Technical Lead, Platform Transition

  • Technical lead for the G-Plans operating-base transition; kept million-plus lead flow moving while legacy and new-user flows overlapped.
  • Supported payment migration toward Go/Temporal, production continuity across funnels, CRM/workflows, internal tools, support, data visibility, Kubernetes/GKE/GitOps services, docs, runbooks, releases, and failure handling.

// G-Plans: first hire to technology lead

Bootstrapped product systems

A long arc of owning product, platform, billing, data, funnels, and production reality under small-team constraints.

Apr 2016
G-Plans

First hire, then full ownership

  • Joined as the first hire and grew into full technology ownership across product, platform, infra, billing, data, funnels, support, and delivery.
  • The operating constraint was small-team reality: architecture had to match cash, headcount, vendor risk, and support capacity.
2018 - 2019
G-Plans

Platform, payments, and production support

  • Hardened the stack into one operating surface: backend, frontend/product workflows, infra/DevOps, production support, release process, internal tools, and billing systems.
  • Built provider-resilient payment behavior with routing, fallback, switching, and risk handling where conventional processors could interrupt or refuse processing.
2020
G-Plans

Funnel economics loop

  • Kept checkout, onboarding, pricing experiments, AOV, early payback, retention, billing behavior, and operational reporting tied together as one revenue loop.
  • Adapted systems repeatedly as the product model, paid acquisition, funnel shape, and first-month retention mechanics changed.
2021 - 2022
G-Plans / Moon

Internal event truth

  • Built a practical data stack around BigQuery, raw events, internal models/layers, attribution continuity, and dbt/Airflow-era analytics engineering patterns.
  • Kept external analytics vendors replaceable by preserving raw ground-truth events and internal reporting continuity.
2023
G-Plans

Compact base for FuturHealth transition

  • The result was not an enterprise platform; it was a compact operating base that could survive pivots, cash constraints, and later handover.
  • That base helped carry early FuturHealth scale and made the later inherited-stack transition possible.

infra

DevOps, deployments, production support, operational control, and small-team maintainability.

funnels

Revenue-critical funnels, checkout, onboarding, pricing experiments, AOV, and retention loops.

payments

Durable billing, provider switching, fallback/routing, payment risk handling, and visibility.

data

BigQuery, raw events, internal models, dbt-style layers, attribution, reporting, and decisions.

Vendor portability

Analytics vendors changed hosting, billing, pricing, and contract shape. The internal event layer protected data integrity and optionality.

Survival economics

Engineering choices were tied directly to acquisition cost, initial AOV, first-month retention, payback, and cash discipline.

FuturHealth base

The result was not an enterprise platform; it was a compact, durable operating stack that could survive pivots and later support handover.

// 2025-2026: agentic development arc

Hands-on agentic tooling

Using LLMs and agents where they make an individual or small senior team more effective: event-based user analysis, Cursor-local triage, E2B sandboxes, Slack-connected adoption, voice input, harnesses, and review surfaces.

2024
Semantic product analysis

Event-based user analysis starts

  • Moved from basic time-series-style analysis into event-based user analysis: each new event updated a current user summary.
  • It worked well for per-user summaries and snippets for support and product people exploring one user, even if it was not yet good enough for aggregated metrics.
Spring/Summer 2025
Moon / event analytics revamp

Spec-driven LLM-assisted rebuild

  • Reworked the Moon event-analytics system with LLM assistance, connecting the older raw-events/attribution thread to newer agentic development practices.
  • The output was useful, but the larger lesson was sharper: without tight context boundaries, harnesses, review surfaces, and generated-code discipline, LLM-assisted rebuilds become hard to keep coherent.
Aug - Sep 2025
Internal operations

Agentic triage helper

  • Built practical Cursor-local triage with internal data connected, so I and then developers/product people could inspect issues, spikes, QA questions, and user behavior.
  • The system pulled together databases, PostHog, communications, product context, payments, CRM, orders, providers, alerts, and distributed failures.
Sep 2025
Cursor / hands-on workflow

Cursor-local triage becomes useful

  • Reached a stable personal workflow for one clear request, one bounded Cursor session, connected data, supervised investigation, review, and iteration.
  • This was the local version before the Go CLI/E2B wrapper made it deployable for people without the same console setup.
Oct - Nov 2025
E2B / ephemeral workers

Deployable Cursor workflow in E2B

  • Wrapped the same local Cursor workflow in a small Go CLI and ran it inside E2B sandboxes.
  • That made the same service available for people who did not have the whole local console setup on their computer, while keeping lifecycle and logs reviewable.
2025 - 2026
Voice input / Mumbli

Speech-to-agent workflow layer

  • Built an initial Qt transcription app in 2025, then revamped it in 2026 into Mumbli, a fast macOS speech-to-text app for turning spoken specs/tasks into agent input.
  • Public repo: github.com/fireharp/mumbli.
2025 - 2026
Wine knowledgebase

Git-driven, source-backed domain repository

  • Built an agent-operable wine knowledgebase for a Valencia store workflow: research new wines, keep citations/sources, run validation checks, and generate site-ready descriptions/details.
  • Explored LangChain/deep-research approaches and ended with a specialized skill-driven repository that can run with Claude, Codex, Hermes, or similar agents; connected WhatsApp to Hermes for store-manager operation.
2026
Multi-agent / software-factory exploration

Multi-lane agents, harnesses, and visual review

  • Exploring multi-agent teams, CI/webhook/event-triggered work, harnesses, context boundaries, verification loops, and structured agent reporting.
  • Built/supports central canvas-like review surfaces where agents show screenshots, working links, structured status, and chart-like output to reduce review load.
Early 2026
Product/business user analysis

Slack-connected user analysis

  • Moved the same direction into Slack: adoption was easier, and Slack brought extra context from the conversations people already had.
  • It was messy until organized around session ledgers and layered/progressive context; then it became more useful for product, support, and organization-wide analysis.

stable

Single-lane agentic development: bounded task, explicit spec, harness, review, iterate.

promising

Triggered maintenance: CI fixes, webhook checks, issue triage, reports, screenshots, recurring operations.

hard

Open-ended multi-lane product work: bottleneck shifts to context switching, review load, drift, and state coherence.

// earlier timeline

Moon, foundations, selected work

Earlier product, web, data, operations, and education context that explains where the current systems instinct came from.

Apr 2022 - Apr 2023
Moon

Co-founder

  • Built product capabilities end-to-end, connected to funnel events, attribution, reporting, and marketing source/channel scale decisions.
  • Continued the G-Plans raw-events/data lineage: preserve ground truth, explain what happened, and turn funnel/acquisition data into decisions.
Aug 2015 - Aug 2016
Initech Software Services Ltd

Full-stack Web and Hybrid Mobile Developer

  • Built web backends and hybrid mobile apps across delivery, deployment, support, and release follow-through.
Sep 2014 - Jul 2015
Bread Software

Front-end Developer

  • Built modular front-end components, HTML5 canvas tooling, testing practices, and maintainable client-side implementation.
2014 - 2016
Web/product shift

Laravel, infinite canvas, full-stack apps

  • Moved into modern product-facing web systems, HTML5 canvas interfaces, front-end systems, and hybrid mobile work.
2010 - 2014
Small business management

Mobile spare-parts distribution, supplier/factory operations

  • Worked around the southern China production and distribution hub during the iPhone repair boom, including suppliers, factories, logistics, and mobile spare-parts distribution across Shenzhen, Guangdong, and Hong Kong.
2009 - 2010
Starlink Telecom (not SpaceX)

Junior Programmer

  • Local provider; university-supervisor-linked role around CORP CRM, Groovy, BPMN, and corporate business-process systems.
2006 - 2012
Moscow Institute of Physics and Technology

BSc, Applied Physics and Mathematics

  • Foundation in calculus, physics/process modeling, numerical thinking, mathematical abstraction, and early neural-network experiments.
1999 - 2005
Programming foundations

MTG TCG in Excel/VBA; Delphi; Borland C++

  • Started with a Magic: The Gathering trading-card-game project in Excel and Visual Basic/VBA, then moved into Delphi and Borland C++.

// selected work

coherence

Go CLI that checks whether code, docs, ADRs, and task references stay coherent before commit.

Mumbli

macOS speech-to-text app for dictation and fast agent workflow input.

knify.dev

Website/product surface for agentic developer workflows and usable project output.

Wine Knowledgebase

Git-driven, source-backed wine catalog/research system for store content operations.

// skills

Backend

Django, FastAPI, Node.js, Go

Frontend

React, Vue, TypeScript

Data / infra

PostgreSQL, MySQL, BigQuery, dbt, Airflow-style orchestration, GCP

Systems

Billing, payment routing, funnels, CRM/workflows, production support, product analytics

AI / agents

Cursor, Claude Code and Codex, MCP-style tooling, Hermes, browser harnesses, E2B workers, specs, review canvases

Education

MIPT BSc Applied Physics and Mathematics