AI & agents
Claude API, Claude Code (skills, headless, sub-agents, hooks), MCP servers, LLM routing (OpenRouter), pgvector / RAG, agent memory, adversarial review agents, spec-driven development (OpenSpec)
I'm Cristhian Fonseca — a Technical Lead with 5+ years building cloud-native FinTech and banking platforms.
Technology serves the business, not the other way around. I build what moves the metric, not what looks good in the codebase.
Right now I lead engineering for a digital credit marketplace inside a major telecom's Super App — ~100K monthly active users within a 45M-user ecosystem. Cloud-native on AWS, automated provisioning, zero-paperwork loans.
I specialize in agentic AI systems — MCP agents, multi-step LLM pipelines, and automation architectures that handle real data volumes with real cost constraints. Every system I publish is first deployed in production.
I believe the best engineers don't just write code — they build systems that compound. And they share what they learn.
May 2025 – Present
Credit marketplace inside a major telecom's Super App
Nov 2024 – Mar 2025
Apps for government-backed regional tourism services
Oct 2022 – Aug 2026
Core banking modules for an enterprise banking client
Apr 2022 – Mar 2023
Cloud services for a US-based car-rental platform
May 2021 – Jun 2022
A global bank's Colombian operation and a telecom operator
Feb 2026
Anthropic
Claude API, Claude Code (skills, headless, sub-agents, hooks), MCP servers, LLM routing (OpenRouter), pgvector / RAG, agent memory, adversarial review agents, spec-driven development (OpenSpec)
AWS CDK, ECS Fargate, Lambda, RDS/PostgreSQL (gp3, Graviton), ElastiCache/Redis, S3, CloudFront, WAF, SQS, Route 53, Cognito, Secrets Manager, CloudWatch, Cost Explorer/Budgets, Terraform, Docker, Caddy, Cloudflare
Python, FastAPI, SQLAlchemy, Alembic, Celery, sqlglot, Java, Spring Boot, Node.js, TypeScript
Angular, ECharts, SvelteKit, Next.js, Thymeleaf, Flutter
PostgreSQL (materialized views, COPY, locking, partial indexes), MySQL, Oracle, MongoDB, Power BI
Sentry, CloudWatch alarms, synthetic probes
Most engineers keep their best architecture decisions private. I think that's backwards. When I ship something that works, I document it and publish it. The community makes me better.
This isn't hypothetical. Every automation template in PowerAI was first deployed in a real system. Every architecture diagram was drawn after the code ran in production.
If you're building with AI, I want to hear what you're solving. The best ideas in this newsletter come from conversations with engineers who are actually shipping.
Hiring for a remote Technical Lead? Email me. Founders and CTOs shipping with AI, say hello too.