01
Skillful AI · my company
Agent platform & workflow engine
The platform layer: agent templates, tool contracts, memory design, and observability that other teams build on.
Reusable agent templates
Built-in observability
Co-founder & CTO · Skillful AI
Agentic AI that survives production.
I architect enterprise agent systems for companies across LATAM, Europe, and North America — regulated data, latency ceilings, legacy mess, and no tolerance for a hallucination when money moves.
Two consulting engagements open per quarter.
Demos are easy. Uptime is the product.
Scale shipped
20k+
messages a month handled by agents in production
Reach
3
continents — LATAM, EU, North America
Experience
10 yrs
ML and platform work — Intel, Apple, then CTO
Verticals
4
car rental, healthcare billing, retail, recruiting
Selected work
Systems running today
Every system here shipped under real constraints — instrumented, documented, still running.
+
02
Lina · fashion retail
WhatsApp lead routing + media AI
Leads routed across 13 stores, B2B wholesale, and e-commerce — with fabric recognition from customer photos and voice notes answered mid-conversation.
~2,000 chats / month
13 locations routed
Problem
A fashion retailer needed intelligent lead routing across 13 store locations, wholesale, and e-commerce — while handling photos and voice notes inside live conversations.
Solution
An assistant that routes each lead to the right WhatsApp channel, identifies fabrics from customer photos with vision models, and transcribes and answers voice messages.
Delivered
Conversation flows & routing policy
Fabric identification (vision)
Voice interaction pipeline
Evaluation + analytics dashboard
FastAPIAWS LambdaWhatsApp APILLM VisionVector DB
03
Talboost · recruiting
AI-powered recruitment platform
Candidate screening, real-time AI feedback, and role-based portals on a multi-tenant architecture.
Real-time AI feedback
Multi-tenant by design
Capabilities
From model to production, one owner
Most AI projects die in the gap between the notebook and the load balancer. I own both ends.
011–4 weeks
LLM applications & agents
Agent workflows, RAG systems, tool integrations, memory patterns, and guardrails on foundation models.
022–6 weeks
Machine learning & analytics
Forecasting, classification, recommendation, and anomaly detection models that decisions actually depend on.
032–4 weeks
Data engineering & pipelines
ETL workflows, lakes, feature stores, orchestration, and the validation that keeps models honest.
041–4 weeks
Cloud architecture & backend
Scalable APIs, serverless patterns, IaC, CI/CD, security hardening, and cost control on AWS.
052–6 weeks
Intelligent automation
Business process AI across WhatsApp, web, email, and API — with explicit logic for when a human takes over.
062–8 weeks
Full-stack delivery
End-to-end product build when AI is one piece of a larger system, database through UI.
Process
Four stages. Each ends in an artifact.
01
Discovery
Goals, users, channels, data, latency and cost ceilings, risk surface.
→ Requirements doc
→ Risk assessment
02
Architecture
System design, tool contracts, data flows, roadmap, acceptance criteria.
→ Architecture diagram
→ Task backlog
03
Implementation
Build, integrate, instrument, and test against failures and edge cases.
→ Working prototype
→ Test coverage
04
Launch
Monitoring, evals, cost controls, and a plan for continuous improvement.
→ Production release
→ Observability plan
Available · 2 engagements, Q3
Costa Rica · all time zones
About
Emanuel Hernández Castillo
I'm co-founder and CTO of Skillful AI, an enterprise AI platform serving automotive, healthcare, and digital commerce clients across three continents.
Before that: GPU software engineering at Intel, data science for App Store and Apple TV+ at Apple, and a decade of ML across growth-stage companies. The through-line is systems that hold up when the demo ends.
LLM agent design and tool contracts
RAG evaluation and grounding strategy
Production FastAPI and AWS delivery
Multimodal pipelines for business channels
Observability, cost controls, reliability testing
Regulated and latency-bound environments
Engagement
Start with one hour
Most engagements start with a single paid hour and a written plan. Scale up only if it's worth it.
Consulting hour
$200 / hour
Architecture review, debugging, or a hard design decision. You leave with notes and an action plan.
Book an hour →Recommended
50-hour block
$10,000 / 50 hrs
Ship a full feature to production. Priority scheduling, async support, weekly syncs, documentation.
Apply via discovery →Discovery
Free / 15 min
A short qualification form, then fifteen minutes to check fit and scope. No pitch.
Start discovery →NDA on request · blocks paid upfront
FAQ
Before you book
What kind of problems are a good fit?
−Where software meets intelligence: agent systems, RAG, ML models, data pipelines, automation, and the backend that carries them. Best fit when the system has to hold up in production under real constraints.
Can you work with our in-house team?
+Advice only, or implementation?
+What do I get after a paid session?
+Which stacks?
+How is scope handled?
+What if it's not a good fit?
+Next step
Book the hour. Leave with a plan.
Bring the architecture problem you've been circling. One hour, written outcome, no retainer required.
Or qualify for the free 15-min call ↓Free discovery
Qualify for a 15-minute call
A few details so I can prep before we talk. High-signal answers beat long ones.