QANTARA · Carlos Salguero
I build LLM pipelines with the engineering most teams skip: grounding, evaluation, quality gates, and a backend that doesn't fall over. Based in Querétaro, working with teams in Mexico and the US.
deterministic extraction · 38 entities · 0 LLM calls
retries 0 · dlq 0 · circuit closed · logged for audit
Documentation turnaround
LLM documentation platform
Referential-integrity score
Zero hallucinated entity names in proof-of-concept testing
Unplanned downtime
Error-ingestion pipeline
CHILECON 2025
Published author · computer vision
Services
Fixed scopes, clear prices, and you work directly with the engineer who builds it.
For teams with an LLM feature that sometimes makes things up, or that nobody fully trusts.
Timeline
1–2 weeks
From
US$3,000
Ship a RAG, multi-agent or document-generation feature that holds up under real traffic.
Timeline
4–10 weeks
From
US$15,000
For retail and service businesses that answer customers on WhatsApp.
Timeline
2–4 weeks
From
US$4,000
+ US$400/mo maintenance
After launch: ongoing support (evals, monitoring, fixes) from US$2,500/month.
Also available: backend and integration work (Go, .NET, Spring Boot, Kafka, Salesforce/IICS) as part of AI projects.
How I work
01
Deterministic extraction before any LLM call, so the model works from facts, not guesses.
02
Eval sets and LLM-as-judge, calibrated against human ratings.
03
Quality scores decide what ships: publish, revise, or escalate to a person.
04
Retries, dead-letter queues, circuit breakers and monitoring that flags problems before customers do.
Work
250+
conversations a day in one pipeline
A bounded Gemini assistant with verified webhook ingestion and idempotent writes, plus a Spanish operator console with human handoff, follow-ups and a compliance log.
Node.js · Fastify · PostgreSQL · Redis
2 days → 2 hours
per document
Sole engineer on a C# modular monolith that grounds generation in tag-based parsing of the source before multi-agent enhancement on AWS Bedrock. Serves ~50 engineers.
C# · AWS Bedrock · Multi-agent · Modular monolith
−60%
unplanned downtime
A real-time pipeline that normalizes those errors into one stream and a dashboard operations teams can act on.
Python · SQL Server · MS Graph API · Power BI
About
I'm Carlos Salguero. I build production LLM systems as an in-house engineer at a Fortune 500 manufacturer, and through QANTARA I bring the same discipline to teams that need AI they can rely on. No account managers, no handoffs.
Accepting new projects
Twenty minutes is enough to tell whether a Reliability Review or a build makes sense.