ALLWELL ONEN

Jul 2026 | Paid client engagement | Ongoing maintenance

• Built an n8n webhook workflow to enrich new HubSpot contacts: fetch company websites, extract context, generate structured company profiles with Gemini, prioritize leads, and write results back to CRM records.
• Client reported sales outreach starting within about 5 minutes of new contact creation, compared with hours or days spent researching companies manually.
• Separated company-site retrieval, LLM-generated enrichment, and rule-based prioritization before updating structured fields in HubSpot.

Apr–May 2026 | Paid client engagement | Maintenance as needed

• Developed a Google Drive-to-Google Sheets screening workflow using PDF text extraction, Gemini scoring against role-defined criteria, and evidence-based justifications.
• Shortened initial screening of 70+ applications from approximately one week to minutes, using consistent criteria defined for the role. Workflow is currently idle between hiring rounds.
• Flagged DOC/DOCX and image-only PDFs for human review when reliable extraction was unavailable, rather than generating unsupported rankings.

Apr 2026 | Paid client engagement | Completed handover

• Integrated Gmail, Gemini, and Google Sheets through n8n to classify incoming emails, record ticket priority and context, and generate brand-appropriate response drafts in the original email thread for human approval.
• For a support inbox receiving 60–80 messages daily, the client reported response turnaround dropping from an 18-hour average to minutes while keeping final-send approval with staff.
• Separated classification and response generation into distinct LLM calls, applying rate-limit pacing and structured-output parsing.

Independent full-stack project | Private development and testing (not client-deployed)

• Building a Next.js/TypeScript application with PostgreSQL, Drizzle, Better Auth, and Cloudflare R2 to extract supplier invoice line items via Gemini/OpenAI and present structured results beside the source PDF.
• Implemented provider-independent model adapters, Zod validation at the AI boundary, OCR-backed source highlighting, and arithmetic reconciliation to flag unreliable invoice values.
• Added tenant-level access checks, transaction-safe trial quota enforcement against concurrent uploads, and automated end-to-end testing across upload, extraction, and review.

Self-initiated case study | Not a commissioned client deployment

• Built n8n automation using QuickBooks, Gmail, Airtable, and Slack: scheduled reminders at days 1/3/7, human escalation at day 14, per-stage logging, retries, and dedicated failure notifications. Documented known rerun and missed-schedule limitations.

Apr 2026–Present | Part-time, Remote | Lagos, Nigeria

• Built Swift Agents Knowledge Base and scheduled/on-demand website traversal features to keep AI customer-support agents supplied with current business content.
• Proposed and delivered hosted/embedded Forms integrations and their documentation; traced a production chatbot-widget failure to obsolete mounting logic and fixed it.

Feb 2023–Apr 2026 | Full-time, Remote | Lagos, Nigeria

• Sole frontend engineer in a five-person SaaS startup, owning React/TypeScript architecture, multi-tenant GraphQL data access, and frontend delivery from inception to launch in three months.
• Built AI-assisted performance review interfaces and productivity workflows as the product evolved to serve 33 organizations with at least 30 employees each.

May 2019–Jan 2023 | Full-time, Remote | São Paulo, Brazil

• Shipped React/Next.js and Vue/Nuxt applications for international clients, including an AI-assisted smart-contract audit interface and localized Bittensor education platform.
• Coordinated server-side pagination for a 4,000+ row dataset; manually observed loading fall from about 5–6 seconds to 1–2 seconds, eliminating page freezes.