About
High-level project management, applied to building software, AI infrastructure and intelligent solutions.
I'm Arnold Torres, an industrial engineer and PMP®, the world's most recognized project-management certification, issued by the Project Management Institute. 6+ years leading technical projects end to end, now applied to building AI infrastructure and software for businesses.
Background.
Start
Project assistant at a biomedical equipment distributor: my first contact with technical projects and vendors.
USA
Freight broker at a carrier company in Los Angeles, California: coordinating operations under pressure and tight deadlines.
2023
PMP® certification from PMI, renewed through PMI-certified continuing education.
Telecom
PM on telecom and software projects: 5G, ASIMOV, drive testing and benchmarking, plus ERP migration, technology migration and software implementation, leading outsourced engineering teams for Huawei, Claro, Entel, Telefónica and LPS (Starlink).
Today
Applying that experience to software development and to implementing AI in business processes, including agents integrated with engineering software: 3D BIM modeling in Revit and lighting calculations in DIALux. Building ARNAI LAB.
Why this matters for your project
Most “AI agencies” are developers who learned to write prompts.
I've spent years leading multi-vendor, multi-country projects with formal methodologies: defined scope, deliverables per phase, a milestone schedule and risk management. That's what I apply to every AI implementation.
It's not “let's try things and see what works”. It's a managed project, with a delivery date and a guarantee behind it.
Certifications.
- PMP® — Project Management Institute (2023)
- Deep Learning Specialization — DeepLearning.AI
- Generative AI for Project Managers — PMI
- Disciplined Agile Essentials + 10 PMI micro-credentials
How I build
Three principles behind every system on this page.
- 01
Safety before the LLM
Anything that must never go wrong runs as deterministic code before the model is called: red-flag triage, message caps, explicit confirmations. The model handles language, not risk.
- 02
Grounded answers only
Assistants answer from a curated corpus or the business’s own data (menu, prices, schedule) and hand off what they don’t know instead of guessing.
- 03
I scope it before I build it
Every build starts with scope, success criteria and risks written down. PMP discipline applied to AI work: fewer surprises and honest timelines.
Toolbox
Capabilities first, tools second.
- Capabilities
- LLM agents with tool use
- RAG / grounded assistants
- guardrails & deterministic triage
- structured outputs
- LLM classification
- Models & AI tooling
- OpenAI / Claude / Gemini APIs
- Whisper (local GPU)
- face detection
- Claude Code + MCP
- Backend & data
- Python
- FastAPI
- Postgres (Supabase)
- SQLite
- Frontend
- Next.js
- TypeScript
- Tailwind
- Integrations
- n8n
- WhatsApp Business API
- Chatwoot
- Telegram Bot API
- webhooks
- Infrastructure
- Docker
- VPS (Easypanel)
- Vercel
- Cloudflare
Open to roles
AI Implementation · Forward Deployed · AI Product.
I bring project-management discipline (PMP) to the messy part of AI: scoping with the business, building the system and staying until it works in production. Remote, from Lima (UTC−5).