Carlos Cruz
Founder & AI Systems Builder · Civil Engineer
Civil engineer and founder. I design, build and operate AI systems in production for real operations — where the physical world meets software.
Open to senior contractor engagements and flexible, asynchronous part-time work up to 30 hours per week; selective remote full-time opportunities considered. Start date to be agreed.
Asunción, Paraguay · Remote
What I built and operate
LICITEX
Public-tender analysis, used inside CQ on real procurement files. An LLM reads the tender documents and matches every requirement against the company profile; a deterministic rules engine makes the final decision and the model cannot override it. A person launches every analysis.
NUVIA
The operations system of Go! Travel, a travel and destination-wedding agency: passengers, events, payments and WhatsApp conversations in one place, on the official WhatsApp Cloud API. An LLM classifies inbound messages, and an internal assistant answers queries about operational data, with every tool bound to the user’s screen permissions.
CONTROLIA
Construction-site control, in production at CQ Ingeniería’s own sites. An AI supervisor inspects twice per working day, reads site photos with vision and cross-checks them against the daily report, the bill of quantities and the contract documents. Anything knowable with SQL is resolved without AI; the model can lower an alert but never create one.
Go! Travel Weddings
Destination-wedding website built for Go! Travel, an external CQ client. Brings together destinations, hotels, galleries and WhatsApp enquiries to help visitors explore the agency’s offer. Published. This commercial website is separate from NUVIA, the same agency’s operations system; no technical integration between them is claimed.
CQ TRINIT
Own CQ product for tax, accounting and financial management with an AI assistant. Implements clients and service concepts, PYG invoices and credit notes, VAT and discounts, atomic numbering, synthetic XML signing and XSD validation, test PDF/QR, audit and sandbox receivables. Private pilot and isolated demo in fiscal simulation. The SIFEN TEST adapter is prepared but disconnected from the fiscal pipeline; official tests, authorisation/homologation and production issuance are pending.
Experience
I design, build, deploy and operate the products above, and I support the people who use them.
I turn an operation into software by working directly with its users — site foremen, an agency team — and taking their feedback into production.
Civil engineer. More than ten years directing construction works, including works for public institutions: teams, clients and public-works contracts.
LICITEX and CONTROLIA come from that first-hand knowledge of tenders and job sites.
Civil Engineering — Universidad Católica Nuestra Señora de la Asunción (UCA), 2012.
How I run it in production
- Queues and jobs. A central job queue with dedicated workers, one exclusive lock per job so two runs never overlap, idempotency keys and reconcilers.
- Azure burst. Heavy render jobs overflow from that queue to ephemeral Azure Container Instances workers — same image digest, pulled from Azure Container Registry — with local fallback.
- LLM engineering. Agents and tool calling, vision on photos and documents, closed-schema outputs with bounded retries, provider-agnostic adapters and per-call cost accounting.
- Stack. Python, Node.js, SQL · PostgreSQL, SQLite · REST APIs, webhooks, WhatsApp Cloud API · Playwright and the Chrome DevTools Protocol · Linux, nginx, systemd.
Additional capability — AI audiovisual production (TRAMIA)
My own end-to-end pipeline — script, scenes, voice, image, video and render — orchestrating 8 generative-AI providers with the cost of every call accounted for.
Languages
Spanish: Native. English: Basic — currently improving. Portuguese: Advanced.
Interested in working together?