Automation & AI engineer
From AI-generated leads to the report that lands on WhatsApp at 7:15 am: I design, build and run systems that take manual work off sales and operations, with measured numbers to back it up.
Automations
This is the real path of a flow in production, with generic labels. Switch to "Failure handling" and watch the backup model and the JSON fixer take over.
A CRM webhook receives the lead, it is enriched with public and credit-bureau data, an LLM returns a schema-validated JSON score, the score is written to the card and to a score table.If the primary model fails or hits its limit, a backup model takes over; if the answer comes back as invalid JSON, an automatic fixer repairs it before validation. A cache avoids paying twice for the same lead.
Work
Corporate projects are anonymous, code-free and use synthetic data. The numbers come from the repositories.
A CRM that fits your process
New field, new pipeline, new automation. No code, no migration, no waiting on IT.
170public API operations
Metadata-driven platform: custom pipelines, visual automations, a documented public API and an MCP server for AI agents. Monorepo with web, worker and decisions recorded as ADRs.
AI lead generation
Describe the market and get leads with verified records, ready for sales, straight into the CRM.
27.4kdeals trained the discard model
Outbound prospecting platform combining generative AI and cascading registry checks: the user describes a market, AI suggests the sectors, every company goes through the same registry rule and the result lands in the CRM in batches.
Self-hosted meeting AI
Decisions and tasks from every meeting, with recordings and history on your own server.
3platforms: Meet, Teams and Zoom
A bot that joins Meet, Teams and Zoom, records, transcribes with speaker labels and extracts decisions and tasks. Swappable transcription and LLM providers, and full-text search across the whole history.
Read the case studySales BI and governance
One rule, from sales rep to the board: the number on the card is the same one the API returns.
13dashboards in production
CRM, SQL Server and PostgreSQL under a single business-rule layer. SSO, per-tab and per-action permissions, daily snapshots and panels for executives and office TVs.
Read the case studyMultimodal AI document triage
The AI reads, classifies and names the customer package, files it in SharePoint and flags what is missing.
94%accuracy on document type
Unzips the customer upload, identifies each document type with multimodal AI (Gemini and Claude), standardizes the name and stores it in the right SharePoint folder, with no manual triage.
Read the case studyBankruptcy filings with OCR and AI
88% less time reading creditor lists, with tax IDs validated and liabilities matching the printed total.
0%liability deviation (was 46%)
Combines native text, OCR and LLMs to extract creditors from PDFs, scans and spreadsheets, validates tax-ID check digits and reconciles liabilities before attaching to the CRM.
Read the case studyMore work
Referral API
Referrals from every source in one API, no duplicates, credited to whoever referred first.
Daily pipeline alert
Every day at 7:15 am, the manager gets the longest-stalled deals on WhatsApp, as a PDF.
Automatic executive report
A weekly report per executive that builds itself from a shared spreadsheet.
Real-time lead triage
The team decides on each lead on screen, live, with the CRM as the source of truth.
Batch contracts
One spreadsheet, one Word template, and every contract comes out as a PDF in one go.
How I work
Every automation starts with a baseline and ends with evidence: logs, counts and dry-runs before writing to production.
Retry only when safe, idempotency, reprocessing queues and alerts. The question is what happens at 3 a.m.
I reuse what exists, avoid needless layers and ship in short cycles with real validation.
LLMs return schema-validated JSON, with model fallback and auditable output.
Describe the process. I will reply with a path to automate it and how to measure the result.
[email protected] · +55 64 9 3300-4125
