Models + toolsWorking stack
OpenAI · Anthropic · Gemini · MCP tools
I use model APIs for reasoning and structured outputs, then connect agents to bounded project context and real tools through MCP.
See the MCP Starter Map →Bogotá · EST-friendly · open to workHey, I'm JP.
Hey, I'm JP.
I design and build useful, reliable AI products.
I'm a Colombian AI Engineer and Product Designer with full-stack range. I shape the experience in Figma and Framer, then use Python, TypeScript, modern model APIs, and production interfaces to ship the system behind it.
JP Casabianca
Bogotá · UTC-5
Open to AI engineering and product design roles where full-stack delivery and design judgment both matter. Remote from Bogotá with strong US East overlap.
Past collaborations
Fuse FinanceLeafLinkCartkitRabinoviciCodesignGrupo InspiroCasabianca CyclingFuse FinanceLeafLinkCartkitRabinoviciCodesignGrupo InspiroCasabianca Cycling
Selected work
A few things I'm proud of.
Each one is a long-scroll case study with the messy middle included — false starts, the prototypes that didn't survive review, the metrics after launch.
8 more in the archive · NDA on 4 of thoseView the full archive →
Capability + evidence
The stack follows the product problem.
I design the experience and engineer the workflow behind it. Tools I use today are separated from the frameworks and infrastructure I'm actively evaluating.
Agent orchestrationExploring
LangChain · LangGraph
I'm comparing chains with stateful graphs, checkpoints, approvals, and recovery paths before choosing a runtime for a given workflow.
Read the runtime comparison →Observability + evalsExploring
Langfuse · LangSmith
I'm testing how traces, datasets, graders, and experiment comparisons can turn AI quality from a demo impression into reviewable evidence.
Open the evaluation starter →Retrieval + vector searchExploring
pgvector · Pinecone
I'm comparing Postgres-native and managed vector retrieval around filtering, answerable context, recall, latency, and operational fit.
Read the RAG evaluation guide →Application deliveryWorking stack
Python · TypeScript
I use Python for APIs, evaluation scripts, and automation, and TypeScript for typed tools, product interfaces, and end-to-end delivery.
See the agent-ready API template →AI product designWorking stack
Figma · Framer
I prototype the human side of AI systems: streaming, tool progress, uncertainty, correction, review, recovery, and accessible fallback states.
Open the AI interface contract →Writing & notes
Things I've figured out, written down.
A working notebook on AI engineering, product design, implementation craft, and the occasional ride outside Bogotá.
FeaturedAug 2026 · 10 min
LLM Eval Confidence Intervals That Matter
Replace dueling percentages with paired prompt differences, bootstrap intervals, cohort checks, a practical-effect floor, and an honest abstain result.
AI Engineering
Read article
Aug
10 min
10 min
Structured Output Validation Beyond Schema
Put generated tool arguments through syntax, supported JSON Schema, domain invariants, and authorization using a hostile mutation corpus.
AI EngineeringTutorial
Aug
10 min
10 min
Agent Token Exchange With RFC 8693
Bind subject, actor, audience, resource, scope, lifetime, and approval into an authority ledger for every delegated tool credential.
AI Engineering
Aug
10 min
10 min
WebNN vs WebGPU for Browser AI
Compare matched model outputs across operator coverage, warm-up, latency tails, memory, fallback, portability, and output parity.
AI Engineering
Aug
10 min
10 min
ConnectRPC vs gRPC-Web for Browser Agents
Replay one agent contract through unary calls, server streams, cancellation, errors, proxies, generated clients, and observability.
Product Engineering
459 posts · RSS available
Tools, gear, stack
What I'm actually using.
Design, engineering, automation, and AI tools I use to move from idea to shipped product.
FigmaReactVueTypeScriptTailwindNode.jsPythonPostmanClaudeCursor
Design
01 / 0401Figma + Framer
UI, systems, and prototypes that feel close to real.
02Adobe CC
Branding, mockups, production assets, and polish.
03FigJam
Research synthesis, flows, workshops, and messy thinking.
04Figma AI
Fast exploration, cleanup, and variant generation.
Frontend
02 / 0401Vue + React
The two product UI stacks I reach for most.
02TypeScript + Vite
Fast iteration with enough structure to scale.
03Next + Nuxt
App frameworks for product teams and content sites.
04Tailwind + PostCSS
Practical styling with room for design systems.
Backend
03 / 0401Node + Python
APIs, automation, scripts, and internal tools.
02Ruby on Rails
Commerce and Shopify-adjacent product work.
03Postgres + Mongo
Relational product data and document workflows.
04Postman
API exploration, testing, and handoff clarity.
AI + Automation
04 / 0401Claude Code
Codebase work, refactors, PRs, and agentic workflows.
02ChatGPT + Gemini
Planning, research, debugging, and second opinions.
03Cursor
Daily AI coding with human review in the loop.
04MCPs
Connecting AI to GitHub, Supabase, Postman, and browser context.
Last updated · May 2026Full /uses page →
Gallery & experiments
The play folder.
Creative experiments, small public builds, and off-screen interests. The things I make to stay sharp.
Small builds
About
AI Engineer and Product Designer—with full-stack range.
I'm Colombian, based in Bogotá, and I work remotely with teams that need someone who can frame the product, design the interface, engineer the AI workflow, and ship the full experience.
I started in Bogotá working across UI, code, and client work before founding Codesign, a studio that mixed design, development, and marketing. That combination stuck: I like being close to the user, the interface, the data, and the business outcome.
My strongest work sits between disciplines: Figma and Framer prototypes, thoughtful AI interaction states, Vue and React frontends, Node and Python backends, API integrations, evaluations, and agent workflows that keep human judgment in the loop.
When I'm not at a screen I'm probably on a gravel bike, playing golf, or testing a new tool to see whether it actually improves the way I work.
Currently · week 22
This week, I'm…
BuildingAI product workflows with model APIs, evaluations, and MCP tools
DesigningHigh-fidelity Figma and Framer prototypes for clear AI interactions
ShippingVue, React, Node, Python, and Supabase-backed apps
LearningLangChain, LangGraph, Langfuse, LangSmith, pgvector, and Pinecone
Quick stats
15+
Years shipping
20+
Projects led
4
Languages
Daily
AI stack
How I work
I design the experience and engineer the system.
The useful part is not just knowing many tools. It is knowing which layer of the product needs attention next—and carrying the decision through to working software.
I start in the product problem, not the file format. That might become a Figma prototype, a Framer flow, a Vue screen, or a small automation that proves the idea faster.
Product design
Figma · Framer · research · systems
Because I can ship the interface myself, I catch constraints early: API shape, loading states, events, permissions, analytics, and the details that make a handoff real.
Engineering
Vue · React · Node · Python · Rails
I approach AI as a product system: model behavior, tool boundaries, evaluations, interface states, human review, and the evidence needed to improve it safely.
AI engineering
Models · agents · evals · MCP tools · AI UX