AIKHEMIST CV

I turn data into gold.

Mauro Orlando

Senior Software Architect Forward Deployed AI Engineer Pre-sales Software Manager

Hands-on senior software and data architect with 25+ years turning business problems, clear or vague, into systems people use in production. Today I design agentic AI systems and run LLM products in production.

  • Remote
  • Full-time, contract or part-time
  • Based in SJC - São Paulo, Brazil (BRT, UTC−3)
  • English, Spanish, Portuguese

2nd place · Dr.DocBench Challenge · EMNLP 2026

See my work Download CV Contact

Highlights

  • 2nd place

    Dr.DocBench Challenge, EMNLP 2026 (oral talk in Budapest)

  • 25+ years

    in software, data and AI architecture

  • First employee in Brazil

    of Cognos and ILOG (both now IBM)

  • AI products in production

    for real users

Enterprise systems delivered for companies such as

How I build systems in the AI era

AI technology changes very fast. A robust architecture has to be ready for that.

  1. Frame: problem before solution.

    I question the request, measure the current process and agree with the client how success will be measured before designing anything.

  2. Map the context: real data is never clean.

    Before any model I map the data, the written and unwritten rules, who decides and what blocks the project.

  3. Design: the business at the center, technology you can swap.

    Business rules sit at the core; models, agents and integrations are adapters around it, drawing on patterns such as hexagonal architecture and domain-driven design.

  4. Validate: measure, don't guess.

    Every AI change is tested on a development set and a locked holdout, and is kept only if it wins on quality and cost. A person approves where errors are costly.

  5. Deliver and learn: from demo to production.

    Monitoring and alerts from day one, the users trained to own the system, and value measured against the baseline.

  6. A method that improves itself.

    Every engagement and every mistake becomes a playbook lesson or a new rule, and a weekly technology radar brings in only what passes the tests.

  7. Build the factory, not only the product.

    AI writes much of the code; I build the system around it: rules files, linters, review agents trained on known bugs, automated tests and human approval.

FDE Engagement Framework

A shared way of working for teams that take complex AI projects to production.

A five-phase method, from a sold project to lasting value, with the client involved at every step.

  1. Frame problem & success

    the client agrees what success means

  2. Map client context

    the client shares rules and data

  3. Design architecture

    the client reviews the design

  4. Validate technology

    the client tests with users

  5. Deliver, close & learn

    the client runs and owns it

12 challenges FDE teams face today

Shared knowledge feeds every phase: a playbook of internal learnings and a technology radar of new research.

Outcomes

  • Client results
  • Lower risk
  • Controlled scope
  • Shared knowledge
The business at the center, technology you can swap

Download the framework (PDF) Click any diagram to see it larger

Selected work

Award

Illustration of document parsing: a warped, scanned page with a tilted table, a formula, chemistry notation and a chart goes through a parser and comes out as clean structure - headings, a rebuilt table, the typeset formula and a redrawn chart.

Self-Learning Agentic Document Parsing System

ArchitectInternational competition

2nd place

in the final standings · 85.83

Solo entry. Invited oral talk and award at EMNLP 2026, Budapest, 29 October 2026.

Problem
Dr.DocBench Challenge, an open international competition at the DocInsights workshop of EMNLP 2026: expert-level document parsing on the pages strong systems still get wrong, turned into structured Markdown with tables, formulas, chemistry notation and reading order intact.
What I did
Built a multi-agent system on LangGraph that turns hard documents into Markdown. Models were chosen per task by cost and quality. One group of agents analyses errors, proposes and tests improvements and keeps only what wins on a locked holdout; another processes each page. Human in command, with approval gates.
Stack
Python, FastAPI, LangGraph, Next.js, LLMs, OCR, document AI

View the final leaderboard

In production

The Kairós site once its opening video ends: the headline over a landscape photograph, a short description and the button to start the journey.

Kairós

Architect & developerCuidar+

Problem
Career transition is slow, generic and hard to navigate.
What I did
Architected and built a hyper-personalized career-transition platform grounded in behavioural science: seven guided phases, 100% of the content created in real time with each user, including about 25 minutes of personalized audio through the ElevenLabs API. Staged LLM agents with model routing and anti-hallucination layers.
Stack
Next.js, TypeScript, Supabase, Claude API, ElevenLabs, HeyGen

Visit site

Inspira

Architect & developerCuidar+

Problem
Leadership learning ran on a paid LMS.
What I did
Built a complete gamified AI learning platform that replaced it: step-by-step learning tracks, simulation games that score leaders, AI avatars, and a real-time dashboard for managers and HR.

Built as Head of Technology at Cuidar+.

The NR-1 diagnostic dashboard: responses, participation, the overall risk index and the eNPS, with a psychosocial risk radar and a bar chart comparing departments.

NR-1 Psychosocial Risk Diagnostic

ArchitectCuidar+

Problem
Companies must map and act on psychosocial risk at work under Brazilian regulation (NR-1).
What I did
Built the diagnostic and analytics dashboards, with risk bands, alerts and eNPS.
Stack
React, Node.js, Railway, analytics dashboards

Visit site

Built as Head of Technology at Cuidar+.

Cuidar+ AI Studio

Architect & developerCuidar+

What I did
Built a shared agent layer whose agents write and publish content to Instagram and LinkedIn after human approval.

Built as Head of Technology at Cuidar+.

The FoodScan Pro app scoring a food: a risk score, a benefit score, the number of ingredients, and lists of risks and benefits.

FoodScan Pro

Architect & developerOwn product

Problem
Consumers cannot read what is really in their food.
What I did
Built a mobile app that scans food labels and returns risk scores, allergy alerts and healthier alternatives, based on WHO, FDA and EFSA references.
Stack
Mobile, computer vision, LLMs

Visit site

Client project

Architecture diagram: client applications call a gateway, which meters usage and credits before routing requests to the computer-vision services.

Kasco AI Store

Backend & frontend architectClient project

Problem
An industrial computer-vision company needed a production backend and store for its AI services.
What I did
Architected and built the store that sells dozens of the client's computer-vision services. Customers send photos and videos through the web UI or a REST API, under a usage-and-credits model.
Stack
Node.js, PostgreSQL, Railway, Playwright

In development

Veradim

Architect & developerOwn product, in development

What I did
An MCP server that gives AI agents a governed BI analyst over company data: semantic layer with approved metrics, data-quality checks, permissions, auditing and lineage. Every generated SQL query is checked before it runs.

View the repository

Cob.Pilot

Architect & developerOwn product, in development

Problem
Critical business rules are locked inside COBOL that nobody on the team can safely change any more.
What I did
Multi-agent pipeline on LangGraph, built against AWS CardDemo: agents read the code, extract the business rules, translate them to a modern language and validate the result against the original behaviour. A human decision screen sits before each critical step, and every agent's work is visible while it runs.
Stack
Python, LangGraph, COBOL, Next.js

Enterprise projects (2002–2009)

Bradesco

BI programCognos (now IBM)

3,000

branches receiving the weekly reports

30+ source systems consolidated into one set of definitions.

Problem
Leadership needed one reliable weekly view of the business, but the numbers lived in more than 30 source systems that did not agree with each other.
What I did
Consolidated 30+ source systems, built a semantic layer so every area read the same definitions, and added a human review step before publication.

Itaú

OptimizationILOG (now IBM)

US$ 1.7M

the third-largest optimization software deal in the Americas

Problem
Armoured cash-transport vehicles had to be routed across a large network under cost, time and security constraints.
What I did
Route optimization for the armoured cash-transport fleet, designed and validated with the client's own analysts.

Petrobras

OptimizationILOG (now IBM)

Problem
Oil-well drilling decisions are expensive, hard to reverse and constrained by what the geology allows.
What I did
Oil-well drilling optimization based on geographic radar data, built with the client's engineering and analysis teams.

Also building

Smaller systems and internal tools, same architecture-to-production approach.

Thymos

In development — Cuidar+

Mobile-first emotional intelligence app that extends in-person workshops into 90 days of personalized practice.

Next.js, Supabase, Railway, Claude API

Athanor

Internal tool — Kairós

Free career diagnostic engine built as a product-led growth channel, with an admin panel, LGPD compliance and cost guardrails.

Next.js, Supabase, Claude API

Reddit Copilot

Internal tool — Kairós

Human-in-the-loop engagement copilot that monitors Reddit, classifies threads and drafts replies under hard rate-limit guardrails.

Node.js, Railway, Supabase, Claude API

Lapistack

In development — Own product

Connects an Instagram Business account, assembles a simulated 9-expert committee and generates a tailored landing page.

Next.js, Supabase, Instagram Graph API, Claude API

Before AI: enterprise foundation

Two decades of optimization, BI and large-scale data work for major companies across Latin America.

Optimization & Decision Systems

Petrobras · Itaú

High-cost operational decisions with many constraints.

  • Advanced route optimization for armored cash-transport vehicles
  • Oil well drilling optimization based on geographic radar data
  • Multiple logistics optimization systems
  • Supply chain

Operations Research · Optimization · Routing · Simulation

Now: Decision Intelligence

Enterprise Business Intelligence

Unilever · Pemex · Codelco · Johnson & Johnson

Leadership needed one reliable view of the business across many systems.

  • End-to-end architecture and development of BI systems
  • ETL, data cleansing and OLAP models
  • Reporting and executive dashboards

Business Intelligence · ETL · Data Cleansing · OLAP · Reporting · Dashboards

Now: Analytics Engineering

Large-Scale Data for Banking & Insurance

Banks and insurance companies

Very large data volumes with hidden patterns and no structure.

  • Processed and organized high-volume financial data
  • Pattern discovery across large datasets
  • Visualization for business and risk teams

Data Processing · Pattern Discovery · Data Visualization

Now: Advanced Analytics · Data Platforms

Enterprise Systems & Integration

Business systems that did not talk to each other.

  • Designed and built enterprise software
  • Integrated ERP and business systems

Software Architecture · ERP · Systems Integration

Now: Solutions Architecture

Brazil · Mexico · Chile · Latin America

What I do

AI Agents & GenAI Products

Multi-agent systems and GenAI applications in production: the architecture, the back end and the interface people actually use.

Built on
Software architecture since the start of my career, plus 2 years at IBM.
Recent work
DocBench Agents (2nd place, EMNLP 2026), Kairós, Kasco AI Store, the Cuidar+ AI agent studio.
Models & tools
Claude, Gemini, ChatGPT, LangGraph, HeyGen (avatar), ElevenLabs (voice)

Legacy & Spreadsheet Modernization

COBOL, Visual Basic and spreadsheets turned into web systems, with the original behaviour tested against the new one.

Built on
My own start: COBOL on IBM mainframes with DB2 for Bradesco and Itaú, then a career spent building systems.
Recent work
Cob.Pilot, a multi-agent pipeline that extracts the business rules from legacy COBOL and validates the translation against the original.

Data for AI, BI & Decision Intelligence

Fragmented data turned into models, pipelines and dashboards that leadership uses — and that an AI system can read.

Built on
8 years in Business Intelligence at Cognos (now IBM), leading project and development teams, plus enterprise ETL/BI projects.
Recent work
Meta Ads performance cockpit, NR-1 psychosocial risk dashboards, Monte Carlo simulation for food companies, investor signal monitoring.
Simulation
Monte Carlo simulation systems: thousands of scenarios generated within company-defined variables, then probabilistic analysis and dashboards showing the most likely outcomes.

AI Evaluation & Guardrails

The layer that makes an AI system safe to ship: anti-hallucination checks, human approval gates, objective metrics and cost limits.

Built on
2 years at ILOG (now IBM) in operations research and simulation, plus the evaluation harnesses I build for my own AI systems.
Recent work
Locked-holdout evaluation in DocBench Agents, per-comment approval queue in Reddit Copilot, human decision screens in Cob.Pilot.
Patterns
Anti-hallucination layers, LLM-provider-agnostic architectures, human-in-the-loop approval channels, cost and rate-limit guardrails.

25+ years from data to AI

  1. 1998–20024 years

    Early career · Origin C&P, HECATON and Axial Informática

    Programmer, then Software Architect & Project Manager

    COBOL on IBM mainframes with DB2: high-volume systems for large banks (Bradesco, Itaú), plus manufacturing and retail. Then a team of 20 on Cognos BI projects, including Mastercard.

    Now: Software / Solutions Architecture

  2. 2002–20075 years

    Cognos (now IBM)

    Pre-Sales & Services Manager, Latin America

    Canadian BI software company (HQ and R&D in Ottawa, now IBM). First Cognos employee in Brazil. Built the Latin America services area with more than 20 partners, hired and led teams of developers and analysts, and worked closely with R&D in Ottawa. Data modernization for Bradesco and for Unilever Latin America.

    Now: Enterprise Data & Analytics Architecture

  3. 2007–20092 years

    ILOG (now IBM)

    Pre-Sales & Services Manager, South America

    Operations research and simulation software. First ILOG employee in Brazil. Optimization projects for Petrobras and Itaú.

    Now: Decision Intelligence · Optimization · Prescriptive analytics

  4. 2009–20112 years

    IBM

    Software Technical Sales Manager (ILOG software)

    Knowledge transfer of operations research and BRMS technology; part of the team behind the third-largest optimization software deal in the Americas.

    Now: Enterprise Data & Analytics Architecture

  5. 2012–20208 years

    Independent Software & Data Architect / Consultant

    Financial, manufacturing and food companies, from problem to production: data layer modernization, decision systems, and Monte Carlo simulations that guided companies on how to operate and what to sell. BI systems, data warehouses and ERP integration.

2020 — present: three parallel tracks

AIKHEMIST

2020–present

AI repositioning since 2021: AI systems architecture and development. Own products and R&D.

Cuidar+

2025–present

Head of Technology

  • Kairós in production since January 2026
  • Thymos in development

Client work

2026

Kasco AI Store — backend & frontend architecture, contract

MBA, FGV

Enterprise systems delivered for companies such as Bradesco · Itaú · Unilever · Johnson & Johnson · Pemex · Codelco · Petrobras · Vale · Cargill

Projects across Latin America

Skills & stack

AI & agents

  • Claude API
  • Gemini API
  • LangGraph
  • MCP
  • tool calling
  • multi-agent orchestration
  • model routing
  • agent memory
  • RAG
  • evaluation with held-out sets
  • human-in-the-loop
  • document AI
  • OCR

Architecture

  • hexagonal architecture
  • domain-driven design
  • FDE Engagement Framework

Software

  • Python
  • FastAPI
  • TypeScript
  • Node.js
  • React
  • Next.js
  • REST APIs
  • COBOL

Data

  • advanced SQL
  • PostgreSQL
  • ETL
  • data warehousing
  • data modeling (star schema)
  • semantic layer
  • IBM DB2

BI, optimization & simulation

  • Cognos BI
  • ILOG (optimization, BRMS)
  • operations research
  • Monte Carlo
  • forecasting

Tools

  • Claude Code
  • Git/GitHub
  • Vercel
  • Railway
  • Supabase
  • Playwright

How I work

  • I work inside the company's team: your engineers, your tools, your repository — not a separate project handed over from outside.
  • Remote-first and async-friendly, with daily overlap with US and European hours.
  • Full-time, contract or part-time.
  • End-to-end ownership, from architecture to production.
  • Clear written communication in English, Spanish and Portuguese.

Outside work In 2011 I reached the summit of Aconcagua, 6,961 m, the highest mountain outside the Himalayas. Fifteen days on the mountain. We started as a group of 21; two of us made the top. The last two days were a blizzard: three metres of visibility, −15 °C. I don't quit.

Raw data is lead. Clear decisions are gold.

EMNLP 2026 · Budapest, Hungary · October 2026

Self-Improving Agentic Document Parsing System

Invited oral talk after finishing 2nd in the Dr.DocBench Challenge, with an overall score of 85.83. The system follows four steps: map every page into blocks, read each block with the cheapest tool that passes a check, assemble everything back in reading order, and learn from every page through an improvement loop with human approval.

Read about the talk

Self-Improving Agentic Document Parsing System: map, read, assemble and learn.
Mauro Orlando

Have a data or AI challenge? Let's talk.

Open to senior roles and projects in AI architecture, forward deployed engineering and applied AI, remote from Brazil.

Download CV