A Slack chatbot with retrieval-augmented generation over thousands of deal documents stored in Dropbox — OCR preprocessing pipeline, vector store ingestion, and grounded answers with source citations. Built for a team that needed to search years of deal memos, term sheets, and diligence reports in seconds.

n8n automations wired into Salesforce for contact and opportunity sync, task generation from email triggers, and automated follow-up sequencing. Custom MCP integrations and a suite of purpose-built AI skills that turn week-long compliance document reviews — contract redlines, signature tracking, regulatory filing status — into minutes.

Also shipped: a custom client document-upload portal with role-based access and investor reporting pipelines. References on request.

A complete internal platform built and operated end to end: Node.js (Express + Vite SSR) application with DocuSign JWT integration for in-app tax-form (W-9) collection and signing — no more chasing paperwork over email. Staff log in, complete and sign forms in-browser, and documents are routed automatically.

Production-grade by default: automatic TLS via Caddy, systemd-managed service with health checks, isolated production credentials, and a one-command deploy pipeline that syncs, builds, restarts, and verifies the live site. Designed, shipped, and maintained by one person.

End-to-end financial modeling for early-stage companies: unit economics analysis, scenario planning, and investor-ready output. We take raw operational data, clean it, model it, and produce the numbers that actually drive decisions — not the ones that look good in a deck.

Built alongside the models: automated data pipelines that pull from live sources (Stripe, QuickBooks, spreadsheets) into a single reporting layer, so the model stays current without manual updates. Founders get a dashboard; investors get a PDF. The underlying logic is auditable and owned by the client.

A full audit of manual processes across a 15-person startup, followed by systematic automation: n8n workflow orchestration connecting Notion, Slack, Google Workspace, and a custom internal API. Eliminated an estimated 20+ hours per week of repetitive work across the ops, sales, and engineering teams.

Deliverables included onboarding automation (account provisioning, tool access, welcome sequences), sales pipeline triggers (lead scoring, CRM updates, follow-up scheduling), and an internal status-reporting bot that surfaces blockers before the Monday standup.

A regulated investment operation needed to redact individual investor data from scanned bank escrow statements — hundreds of pages, degraded scan quality, zero tolerance for leaking the wrong name. Off-the-shelf OCR failed. So we built a dual-engine pipeline: multi-pass Tesseract that runs several preprocessing and config combinations and scores each candidate, backed by a deep-learning fallback (docTR — neural text detection + CRNN recognition) for pages classical OCR can’t read.

Deployed as a production service on a hardened VPS. What used to be hours of manual, error-prone redaction per investor is now a one-command operation.

We run our own AI agents on our own metal — not someone else’s cloud dashboard. A persistent agent deployed on a hardened European VPS: SSH bound to a private mesh VPN only, public access disabled, deny-all firewall, fail2ban, rootless containers, and integrity-hashed identity files so any tampering with the agent’s instructions is detectable.

The agent handles real work — document triage, research, Slack — under strict human-approval boundaries. Privacy and control aren’t features we bolt on; they’re the architecture.