Norrköping, Sweden · Available for work

I build the systems that run the business.

Ten automation products in production for retail, logistics and hospitality — voice agents, retrieval-augmented chatbots, invoice OCR pipelines and operational dashboards. Designed, built, deployed and maintained solo.

See the work Download CV (PDF)
Mario Soriano Bañuls, automation and AI developer
10Automation products in production
2Live client deployments
159Automated tests · 8 CI pipelines
274Screens under visual regression

01 — Capabilities

What I build

Four areas. In each one I have shipped something that a business depends on every day, not a demo.

01

Business software

Operational platforms with financial control, KPI dashboards, supplier management and a client-facing portal. Multi-tenant, with role-based access and an audit trail on every mutation.

02

AI systems

Retrieval-augmented chatbots over Supabase with vector search, AI voice agents that book and confirm, OCR pipelines that extract structured fields from scanned documents, and natural-language-to-SQL for analytics.

03

Automation pipelines

Orchestration with n8n, webhooks and REST APIs. Website lead forms validated, scored, written to a database and surfaced on an internal dashboard — with retries, rate limiting and dead-letter handling.

04

Programmatic video

Brand and product video produced as code with Remotion — reusable motion components, versioned renders, no editor and no designer dependency.

02 — Selected work

Two systems in production

Client names withheld. Screenshots are the real production interfaces, with the internal support banner and the sidebar removed.

Operations dashboard showing 42.40 kg sold, 1073.15 euro cost, 1323.77 euro revenue and 250.62 euro margin, with a species margin breakdown, per-supplier margin chart and a purchase-versus-sale price scatter plot

Margin control and sales pricing for a day of landed fish — species breakdown, supplier margin and a price-point suggestion engine.

Case 01

Operations platform for a fishing cofradía

Alicante, Spain · live client deployment · built and maintained solo

Delivery notes arrive by email as scans. OCR extracts the supplier, weight, species and price; each note is then cross-validated against that supplier's own history to catch duplicates, unusual amounts and unregistered tax IDs.

The result feeds per-species margin control, supplier management, sales pricing and a client portal. A pricing engine computes a suggested price per species from a target-margin formula and flags every line that sits below it.

12Admin routes
11Client portal routes
57Components · 55 unit tests
161 KBServer-side logic
Live voice-agent dashboard showing a restaurant floor plan with 22 tables coloured by availability, occupancy metrics and a real-time reservation panel

Voice agent control panel — floor state, table availability and reservations updating live.

Case 02

Automated ordering with an after-hours voice agent

Agricultural supplier, Spain · live client deployment

Three channels, one system. WhatsApp order capture handled automatically, a voice agent that answers calls outside business hours and books or quotes, and a web assistant that routes visitors either to purchase or to the grower information area.

Everything funnels into the same backend: leads scored, stored, and pushed to an internal dashboard, so the team reviews the pipeline rather than reading forwarded emails.

3Automated channels
24/7Calls and messages handled

03 — Programmatic video

A brand trailer, produced as code

Most teams need a product video and have no designer and no motion budget. This one is written in React and rendered by a build pipeline.

Play the 59-second trailer 58.9s · 1920×1080 · rendered on self-hosted infrastructure

Loads nothing from YouTube until you press play — no third-party request on page load.

58.9s1766 frames at 30 fps, 7 scenes
39Video compositions in the library
258Reusable motion components
86,691Lines of the video codebase

The audio is not a single track either: eleven frame-accurate ducking zones let the music drop under each spoken line, and the master is normalised to −20 LUFS. Renders run on a self-hosted Docker server tagged with immutable version strings, so any past video can be rebuilt byte-identically.

Why this matters outside marketing: I can produce the video asset a product needs without a designer and without After Effects — re-runnable, versioned and diffable in git.

04 — Engineering practice

The parts that do not show up in a screenshot

Every number below is a count of files or pipelines in the repositories behind the work above.

741
TypeScript / React source files
23
Database migrations applied
15
API routes and webhooks
159
Automated test files
8
CI pipelines, incl. visual regression
10
Archived Lighthouse audits
TypeScriptReactNext.js Pythonn8nSupabase PostgreSQLREST APIsRemotion DockerCloudflarePlaywright VitestGitHub Actions

I run the infrastructure myself: self-hosted Supabase (Postgres, Auth, Edge Functions), n8n for orchestration, Docker and Coolify for deployment, Cloudflare at the edge. Deploys are gated by CI, and a Cloudflare cache purge runs automatically after every release.

Invoice analytics panel showing spend totals, VAT, top suppliers and monthly tax breakdown
Invoice pipeline — OCR extraction, duplicate and fraud checks, mapped straight into an ERP.
Lead pipeline dashboard with scores, volumes sent, reply rate and stage counts
B2B pipeline — scrape, validate, score, send, follow up. Only aggregate numbers shown.
System health dashboard showing workflow status, VPS metrics, Redis state and error history
Observability — n8n workflows, VPS metrics, SMTP reputation and error history in one place.

05 — Experience

Where the work happened

2026 — present
Babieca Tech
Co-Founder & Software / Automation Developer
  • Co-founded an AI and automation company serving small businesses, covering software development, process analysis, marketing and client delivery.
  • Designed, built and operate a suite of ten automation products currently running in production, two of them as live client deployments.
  • Reduced a recurring 45-minute daily manual process to an automated workflow that now only needs review of the resulting information.
  • Use LLMs as a development tool throughout: code generation, debugging, architecture exploration, testing, documentation and prototyping.
2024 — 2025
RockStar Data
Data Analyst
  • Built ETL workflows integrating customer data from CRM, ERP and POS systems, with automated validation to improve reliability and reduce data-quality incidents.
  • Defined, calculated and reported KPIs supporting pricing and cost optimisation decisions.
2026 — 2028
Linköpings universitet
MSc Computer Social Science
2019 — 2023
University of Alicante
BSc Sociology

06 — Contact

Open to work in and around Norrköping

Native Spanish speaker, English at IELTS Academic C1. Available for data, AI, automation and software roles — remote or on site.

datamario.ba@gmail.com LinkedIn GitHub

CV — pick the relevant one Data, AI & Automation PDF Computer Social Science PDF International / part-time roles PDF