Work samples

Private AI workflow samples.

A live retrieval demo, search build notes, and practical patterns for local AIOS layers, tool-using workflows, routing, support, reporting, and dashboard systems.

Live retrieval demo Private search notes Local AIOS patterns
OLLAMA / VLLM
PGVECTOR
HUBSPOT
PRIVATE VPC
N8N
Start here
Search build notes

ScoutLocal: AI search for messy marketplace queries

Next.js 15 • NestJS • Prisma • pgvector • private inference-ready architecture • Clerk
Status: build notes

Business problem

Marketplace search breaks when users type vague human queries like "vibey coffee spot" instead of exact categories. The goal was to make search more forgiving without letting AI invent venues or ignore real filters like location and hours.

What Was Built

  • Hybrid search: Semantic matching for natural-language intent, anchored by SQL filters for facts the system cannot guess.
  • Cost control: Debounced map fetching to reduce wasted API calls from mobile pan/scroll behavior.
  • Maintainable data model: Clerk + Prisma identity patterns for merchants that can appear across different marketplace roles.
Natural-language search pattern
Latency-aware mobile architecture
Shopify GraphQL Integration

Fulfilly: Multi-Vendor Shopify Transaction & 3PL Fulfillment Sync

Shopify Admin GraphQL API • Supabase • Node.js • Webhook handlers • PostgreSQL
Status: production build

Business problem

Mid-market e-commerce merchants lose hours manually syncing shipping statuses when they source products across multiple third-party logistics (3PL) warehouses. Mismatched carrier tracking numbers lead to customer support backlogs and duplicate fulfillment states.

What Was Built

  • Shopify Admin GraphQL: Queries nested `fulfillmentOrders` by invoice parameters to target precise line items across multi-warehouse splits.
  • Transactional Mutations: Executes secure `fulfillmentCreateV2` mutations to programmatically bind tracking details and trigger native carrier notifications.
  • Relational Logging: Hosted on Supabase to log incoming client webhooks, record payload changes, and run robust retry cycles to prevent duplicate transactions.
Shopify GraphQL mutations
Multi-vendor fulfillment sync
Serverless Integration

ScheduleAI: Voice-Based Autonomous Appointment Booking Agent

Twilio Voice API • OpenAI API (Structured Outputs) • Serverless Functions • Relational Database
Status: production build

Business problem

Small businesses lose up to 30% of booking conversions when phone inquiries arrive outside working hours. Rigid automated interactive voice response (IVR) flows lead to high drop-off rates and empty calendars.

What Was Built

  • Twilio Audio Stream Webhooks: Stream incoming audio files over secure websockets to a serverless node infrastructure for instant, low-latency processing.
  • Grounded Calendar Tooling: Uses OpenAI structured tool calling to parse spoken times, validate schedule boundaries, and query database availability.
  • Conflict-Free DB Writes: Locks calendar times and executes atomic SQL transactions to save booked sessions, preventing double-bookings.
Twilio Voice Webhooks
OpenAI Tool Calling
Other workflows I build

Automated Lead Routing

Inbound leads sit in forms, inboxes, spreadsheets, or CRMs while someone manually qualifies and assigns them.

→ Faster response to qualified leads

Invoice Automation

Closed-won deals still require someone to create invoices, check payments, and update finance tools by hand.

→ Less founder/admin bookkeeping

Support Intelligence

Support teams repeat the same answers and lose time deciding where each ticket should go.

→ Drafts and routing with human review

AIOS / Agentic Systems

Teams need an operating layer where AI can understand context, use approved tools, remember workflow state, ask for approval, and leave an audit trail.

→ Context, tools, memory, controls
View AIOS systems

Ops Dashboards

Teams need a shared view of workflow status, exceptions, handoffs, and what the controlled system already handled.

→ Review queues and live operating metrics

Legacy Database Sync

Connecting legacy software or local databases to modern cloud tools usually requires brittle, custom middleware that breaks without warning.

→ Robust, monitored data pipelines

Have a workflow worth testing?

Send one bottleneck. I will tell you whether it needs a no-code fix, a focused system repair, or a scoped AI build.