Systems and integrations developer
Alex Henrique da Silva
I do systems integration and process automation inside manufacturing ERP.
I have been at Adapcon since 2021, in southern Brazil, developing on Consistem ERP (InterSystems IRIS / ObjectScript). The work almost always starts the same way: there is a process that only happens because someone remembers to do it, and it needs to start happening on its own, with a deadline it has to meet and a record of who did what. To get there I connect the ERP to whatever the company runs around it: B2B marketplaces, e-commerce, store tills, WhatsApp, tax robots, cloud services, AI models. When the process needs to leave the ERP, I also write the web app or PWA that talks to it. The case studies below are anonymised. I talk about industry, scale and architecture, with no client names and no source code.
By the numbers
590+
signed routines and classes
48
repositories
32
clients and products served
5
years of experience
Recent highlights
- Payment receipts sent over WhatsApp, read by AI (OpenAI and Claude) and reconciled against the order inside the ERP, all the way to settling the receivable.
- Triangular invoicing across group companies with idempotent step control: reprocessing never duplicates a stock movement.
- A multi-step workflow engine built from scratch in 2026, with per-step deadlines and automatic overdue alerts.
- Retroactive IBS/CBS tax recalculation on already-issued invoices, ready within the deadline of the Brazilian tax reform.
- Features of mine in the core ERP product: an implementation panel with bulk imports, and the employees REST API.
Selected case studies
Real areas of work, told through industry, scale and architecture.
B2B e-procurement engine for pharmaceutical distribution
A multi-marketplace quoting engine integrated with Bionexo and Sintese, plus catalogue and order synchronisation to the field sales platform over REST and WebSocket.
Payment receipts over WhatsApp, read by AI and settled in the ERP
The receipt arrives on WhatsApp, an AI model extracts the data into a fixed format and the ERP reconciles it against the order through to settlement. Idempotent webhook, two clients, two AI providers.
REST integration with a retail POS and Brazilian tax reform rollout
A two-way bridge between the ERP and the Varejo Online POS platform, with a custom rate limiter, log-driven incremental sync and retroactive IBS/CBS tax recalculation.
Hybrid tax and banking RPA, from Power Automate to an in-house API on AWS
Automation of tax obligations and banking routines with the rules inside the ERP and execution outside it, in Power Automate and an in-house API on AWS, with retries and observability.
Triangular invoicing across group companies, safe to reprocess
The order starts as raw material, goes through cutting and is invoiced as a finished product by another company in the group. Idempotent step control and cascading reversal when an invoice is cancelled.
WhatsApp-based sales CRM with RFV segmentation
Backend for a sales CRM app with a 360º account view, RFV segmentation, visit logging through a WhatsApp bot with audio transcription and automated NPS surveys.
Offline PWA and public pages connected to the ERP
Front-ends outside the ERP: an offline inventory PWA for an Android handheld and public pages where the customer approves orders and samples through a link, all talking to the ERP's API.
Multi-step workflow engine built from scratch
A generic workflow engine inside the ERP: step definition, rule-driven progression, side effects on completion and an SLA job that flags overdue work.
Onboarding console shipped in the core ERP product
A core-product ERP feature for bulk-importing master data during client onboarding, with a tax authority lookup later extracted into a reusable component.
End-to-end acceptance testing with AI agents
An in-house tool that builds the test script from what changed in the code, runs it on the client's acceptance system through Playwright and records expected vs actual live, with a fixer agent for failures.
E-commerce, marketplaces and live commerce
Digital sales integrations on top of the ERP: Tray, Bling, a marketplace hub, drop-shipping with an external franchise, live commerce and a two-way textile PLM bridge.
WMS with regulated batch traceability
Warehouse management for pharmaceutical distribution: batch traceability, picking labels, pending-stock reconciliation and cascading release of commercial and credit holds.
Experience
Adapcon
May 2021 – presentJaraguá do Sul, BrazilMid-level software developer
Feb 2022 – present- ERP customisations and integrations for 32 clients and products, across textiles and apparel, pharmaceutical distribution and metalworking.
- Integrations with B2B marketplaces, e-commerce, store POS, TMS, WhatsApp, AWS services and tax and banking RPA.
- AI automation: payment receipts read by OpenAI and Claude, reconciled against the order all the way to settling the receivable.
- Architect and main author of a generic workflow engine inside the ERP (2026).
- Retroactive IBS/CBS tax recalculation on already-issued invoices, delivered within the deadline of the Brazilian tax reform.
- Features in the core ERP product: an implementation panel with bulk imports and the employees REST API.
- Front-ends outside the ERP: an offline inventory PWA and public approval pages in React and TypeScript.
- In-house end-to-end acceptance testing tool built on AI agents and Playwright, adopted by the team.
Junior programmer
Jul 2021 – Feb 2022Trainee programmer
Jun 2021 – Jul 2021Intern
May 2021 – Jun 2021
Technologies and integrations
Platform
Applied AI
Cloud and automation
Front-end and testing
Messaging and support
Digital commerce
B2B and healthcare
Tax and compliance
Productivity and HR
How I work
Business rules stay in the ERP
Robots and integration scripts execute. The ERP decides. The moment a rule leaks into external automation it stops being auditable and starts breaking every time the vendor touches a screen.
I treat silent failure as a bug
Every integration I ship comes with logging, alerting and a panel where you can see which step the process stalled at. The worst version of a problem is the one the client finds before I do.
When the request is a symptom
If a ticket is the third variation of the same problem, I stop and build the generic mechanism instead. That is what happened with the workflow engine and with the marketplace adapter layer: rather than one more bespoke screen, a mechanism that handles the next cases through configuration.
Reuse the validation that already exists
Bulk imports and integrations write through the same routines the official screens use. Writing straight to the database saves you an afternoon and charges you later, when the rule changes in only one of the two places.
Reprocessing has to be safe
Integrations drop, users double-click, jobs run again. Every flow that touches stock or finance I design to resume where it stopped without redoing what it already did. And a data fix in production only ships with a dry run, a backup and a scope checked against the live data first.
AI speeds me up, I answer for the result
I use AI agents to specify, implement in parallel and test. But the model returns data in a format with a contract, the rule stays in code, and a person signs off before anything reaches the client.
Contact
I am currently employed, but I am always up for a no-strings conversation about systems integration and process automation. WhatsApp gets the fastest reply.