Est. MMXXIV · Hermosillo, MX 29.0729° N · 110.9559° W
Available Q3 2026

Iván
Martínez
Agüero.

Operator and builder. Fifteen years shipping production systems. Now building AI that actually runs.

[ photo ]
15+yrshipping
40+engagements
No.001

Now.

Workshop log · Updated weekly
01BuildingAn infrastructure provisioning agent — natural language to live Fly.io stacks in under 90s. Hard part is validation before apply.May 2026
02LeadingTechnical lead at EqualsTrue — system design and AI lab. Multi-agent eval harness, ADR-driven architecture governance.Ongoing
03ShippedModel router v2 — privacy overrides, per-workload routing, fallback chains. 40K+ req/day at <12ms routing overhead.Apr 2026
04ReadingDesigning Data-Intensive Applications ch. 11 · Anthropic interpretability papers · Anyscale on LLM serving economics.This month
05WritingNew note — The $50K AI budget mistake. Demo-to-production mismatch. Architecture conversation that should happen in week one.In draft
No.002

Selected work.

40+ engagements total
001

Radian Corporation — AI Pricing Engine

Previous POCs collapsed at concurrency. Rebuilt the infra layer — routing, caching, eval harness, observability on day one. Production from week one.

InsuranceFastAPIClaude APIpgvector
500msp95 latency
94%accuracy
99.8%uptime

Context

HomeGenius needed a high-throughput pricing API serving thousands of simultaneous broker requests — each requiring real-time home value calculations across multiple risk and market factors.

What I built

Rebuilt the infra layer from scratch: async request routing over AWS SQS to decouple broker intake from model inference, pgvector-backed semantic caching to skip redundant calculations on similar inputs, and a factor-based pricing engine that accounts for property characteristics, location signals, and market conditions. Full observability from day one.

Stack

FastAPI · Claude API · AWS SQS · pgvector · PostgreSQL · CloudWatch

002

EqualsTrue — AI Lab & System Design

Technical lead for system design and the AI lab. Owns ADR process, LLM routing layer, eval pipeline, and multi-agent prototyping for the platform.

Technical LeadMulti-AgentLangGraphOngoing
2yr+embedded
40K+req/day routed

Context

AI-powered recruitment platform scaling faster than its original architecture could support. Needed both a technical foundation and an internal AI capability — not just features shipped, but the systems thinking behind them.

What I built

Own the system design for internal workflows: ADR process, service boundaries, async patterns, and data contracts. Separately run the AI lab — LLM routing layer, eval harness, and multi-agent prototypes using LangGraph. Also lead infra decisions: NestJS API, Prisma/PostgreSQL, AWS EC2, Docker, CI/CD pipelines.

Stack

NestJS · TypeScript · PostgreSQL · Prisma · LangGraph · Claude API · AWS EC2 · Docker · GitHub Actions

003

SaaS Platform — AI Infra Provisioning

Natural language to live Fly.io + Docker Compose stacks in under 90 seconds. Validation layer before apply. Self-healing loop handles 94% of runtime issues autonomously.

InfrastructureFly.ion8nAgents
<90sprovisioning
−87%devops hours

Context

An AI-powered cloud platform for infrastructure provisioning. Teams describe what they need in plain language — the system handles the rest, from config generation to deployment to self-healing.

What I built

Natural language input → LLM-generated cloud configs → validation layer (dry-run + schema check) → live deployment in under 90 seconds. A self-healing agent monitors runtime health and resolves ~94% of issues autonomously — restarts, config drift, resource limits — without human intervention.

Stack

Fly.io · Docker Compose · n8n · Claude API · GitHub Actions

004

Shun Technologies — Unified Vendor Experience

Unified chatbot and workflow platform for supply chain operations. Bottler and mounting automation, ML-driven demand signals, and vendor management in a single interface.

Supply ChainMLChatbotWorkflows
−18%stockouts
−23%holding cost

Context

Supply chain operations spread across disconnected tools — vendors, bottlers, and mounting operations each with their own systems. No single view of demand, inventory, or supplier status.

What I built

Unified vendor experience platform: a conversational interface for operations teams to query inventory, trigger workflows, and surface ML-driven demand signals. Bottler and mounting automation reduced manual coordination overhead significantly. Integrated directly with WMS and ERP systems.

Stack

Python · ML ensemble · LLM chatbot · REST · PostgreSQL · WMS/ERP connectors

005

UniTravel Tech — Dynamic Pricing for Tourism

Sub-100ms pricing API for 50K+ daily requests. ML demand forecasting, competitor rate signals, and yield management across 12 LATAM tourism markets.

TravelTourismReal-Time PricingYield Management
+14%revenue lift
<100msp99 latency

Context

Tourism marketplace competing across 12 LATAM markets where pricing windows are narrow and competitor rates shift intraday. Static pricing was leaving margin on the table.

What I built

Sub-100ms pricing API backed by a caching layer over ML demand forecasts. Competitor rate feeds drove a yield management model adjusting prices every 15 minutes per market. Rolled out market-by-market with an A/B harness to validate lift before full deploy.

Stack

FastAPI · Redis · ML ensemble · PostgreSQL · competitor rate feeds · AWS

No.003

Notes & field reports.

Production-first · One per week
No.004

Building in public.

Open source · Tools I use daily
No.005

Newsletter.

Field notes · Every Friday
Systec Field Notes

One note per week. Real problems, real code.

Not a newsletter about AI in general. A dispatch from inside production systems — what broke, what shipped, what I'd do differently. Operators and founders only.

No noise. Unsubscribe any time. Sent every Friday.
✓ TRANSMITTED

You're in. First issue Friday.

Check your inbox for a confirmation — sometimes lands in spam.
4Issues out
~5minAvg read
FriEvery week
Latest issue
The model router problem — privacy, fallback, and trust
No.006

Hire me.

Direct · Honest · Fast intake

Two paths in. One conversation either way — no pitch decks, no discovery forms.

[ photo ]
Iván Martínez Agüero
Operator · Builder · Fractional CTO

If you have a process to automate: start with a Readiness Sprint. Two weeks, written report, fee credited toward whatever ships next.

If you're earlier — exploring, evaluating, gut-checking — just write. One paragraph. Forty-eight hour reply.

Path AAI Readiness SprintTwo weeks · Report · Debrief.
Path BJust writeOne paragraph · 48-hour reply.
Path CFractional CTOEmbedded · ADR process · AI lab.
AgenciesPartner programWhite-label delivery · Request brief.
Built by Iván.
Hermosillo · Sonora · México · MST