ANDRÉ TAKEO FUJIWARA
Forward Deployed AI Engineer working embedded with clients to ship AI agents into critical, high-volume processes. At Adalink I'm the sole developer of four agent-based systems for Amcor, a global packaging multinational — across pricing, master data, tax compliance and payroll. I work in RAG, agent engineering, MCP and Claude Code, with a bias for human-in-the-loop and auditability.
01 / STACK
SKILLS & STACK
The tools I reach for — chosen for reliability and speed, and because I know them deeply.
- React · Next.js
- TypeScript
- Tailwind CSS
- Streaming UX
- Node.js
- NestJS
- Python
- REST APIs
- RAG · Qdrant
- Agent Engineering
- MCP · Claude Code
- LLM Ops · AI Ops
- PostgreSQL · MongoDB
- Redis
- Docker
- Grafana · Prometheus · OTel
02 / WORK
AI AGENT SYSTEMS
Four agent-based systems I built solo for Amcor — a global packaging multinational — applied to critical, high-volume manual processes. Every one is human-in-the-loop and auditable.
RFQ & Pre-Quote Consolidator
Unified pre-quote data scattered across spreadsheets, CRM and production systems. A multi-source ETL pipeline with an AI agent doing semantic field normalization — structure names, units, product codes — and flagging incomplete records for human review. Unlocked margin analysis and RFQ-to-product traceability that was impossible without manual search.
TypeScript · Next.js · AI agents · ETL
Master Data Sanitizer
AI-assisted data-quality system for critical master bases — materials, prices, accounting — that block downstream processes. Deterministic rules plus fuzzy matching and an AI agent suggesting fixes for duplicates, inconsistent codes and missing fields. Strictly human-in-the-loop: every suggestion is approved before any mutation, with a full audit trail.
AI agent · fuzzy matching · approval workflow
Fiscal Compliance Robot
An AI agent that compares monthly tax filings against parameterized rules (ICMS-ST, PIS/COFINS first) and flags suspicious lines — atypical rates, divergent bases, ERP mismatches — each with an auditable justification carrying the violated rule and its reasoning. Rules live in a parametric layer, so legislation changes need no code.
AI agent · parametric rules engine · line-by-line validation
Payroll & Benefits Audit Robot
A validation engine cross-checking payroll events against benefit-eligibility rules — leaves, meal vouchers, pharmacy program — surfacing suspicious payments before monthly close. A parametric rules engine lets HR encode rules without dev involvement, and an AI agent generates business-language explanations for each inconsistency. Acts on one of the company's largest cost centers.
parametric rules engine · explanatory AI agent
03 / EXPERIENCE
EXPERIENCE
Where I've worked and what I shipped — a few highlights from the road so far.
Adalink · Barueri, SP · Hybrid
Sole developer of four AI-agent systems delivered to Amcor, a global packaging multinational — applying agents to critical, high-volume manual processes across Pricing, Master Data, Tax and HR, always human-in-the-loop and auditable.
TPF Engenharia · São Paulo
Automated reporting pipelines in Python and built operational Power BI dashboards. Refactored and optimized long SQL Server queries, significantly cutting report generation time.
Opus Software · Remote
Built full-stack internal tools with React.js and delivered client-facing features, working across the stack with Git-based workflows.
PUC-SP · Exchange at KU Eichstätt-Ingolstadt
Finishing my Bachelor's in Data Science & AI at PUC-SP, with an AI-focused exchange year at KU Eichstätt-Ingolstadt in Germany. Languages: Portuguese (native), English (C1), German (A2).