Automations for various ventures.

A per-section pipeline that turns raw neuropsychological assessments and clinician transcripts into structured, deidentified reports — with evals at every step and a human in the loop wherever the model isn't sure.

• Ingest & Deidentify: Raw assessments and transcripts come in via OCR and the Extend API, get deidentified with Presidio, and land as a normalized JSON document per case
• Section Pipeline: A dedicated sub-pipeline per report section — each runs extraction, standardization, and formatting against that section's schema
• Step-Level Evals: LLM-as-judge against a mix of distilled and raw context for grounding, plus deterministic checks on formatting and required-field presence — every step is scored before it can advance
• HITL Dashboard: Surfaces each step's output and flags low-confidence spans for clinician review before anything leaves the system

Operations and revenue-side automations for a logistics business — pulling four separate systems into one dashboard, tracking procurement state through email instead of rigid status updates, and feeding cold leads in from the open web.

• Unified Ops Dashboard: Internal dashboards that pull ServiceCore (field service), SAP (ERP), QuickBooks (accounting), and Extensiv (3PL / warehouse) into a single operational view — one place to see order, fulfillment, and financial state side-by-side instead of swiveling between four systems
• Procurement Tracker: Mailgun ingests every email CC'd to the pipeline mailbox, OpenAI classifies each thread into a procurement stage (inquiry → quote → PO in SAP → invoice in SAP → shipping confirmation → delivered in Extensiv → payment reconciled in QuickBooks), and the oversight dashboard reconciles AI-inferred state against the systems of record so any drift surfaces immediately
• Cold-Lead Enrichment: Scrapes target sources through Bright Data, dedupes against the existing book, and pushes structured contacts into HubSpot and Salesforce with first-touch metadata pre-filled

Built in close partnership with the CFO and CEO of a company managing the web presence of 10,000 medical practices. Three workflows aimed at the parts of the revenue org that were eating the most human time — exec reporting, ticket triage, and SDR prospecting.

• Exec Sales Reports: Pulls live numbers from Salesforce, joins context from Google Sheets and Docs, and ships a formatted brief into the executive's email or Slack (DM or channel) on a schedule. Replaced hours of weekly hand-rolled reporting for the CEO
• Ticket Classifier: LLM-driven Zendesk ticket router that categorizes inbound tickets by language required, estimated skill level, and current agent backlog, then assigns or recommends. Low-confidence cases defer to a human dispatcher instead of being auto-assigned
• SDR Lookup Extension: Browser extension for SDRs working cold lists — when the rep lands on a prospect's site, the extension looks the company up in Salesforce in the background. If it exists, last-contact context appears inline; if not, a quick-input form pre-fills a new record. Cuts the lookup-or-create cycle from minutes to seconds

Automations for content teams shipping across many channels — turning a single brief into coordinated assets with brand-voice constraints baked in.

• Brief-to-Draft: Takes a structured brief and produces blog, social, and email drafts that share a single voice and CTA
• Asset Orchestrator: Ties image and video generation to the copy, gated on a brand-review step before anything is published
• Performance Loop: Pulls post-launch metrics and feeds them back as constraints on the next brief — what worked, what to avoid