Shipped software, not slideware. With the receipts.

ProblemProperty managers needed programmatic and AI-driven access to an HOA/community-management system of record — safely, without runaway automations quietly hammering the API.

BuildA Standard-API MCP server on a hardened cloud host (systemd + secure tunnel), a manager-to-association mapping tool, connectivity into workflow automation and an AI connector, and Datadog monitors that alert on runaway loops — surfaced back in an operator control view.

ResultA monitored, production MCP endpoint that lets both automations and AI assistants act against the platform with guardrails and observability.

• check_circleMCP server running in production behind a secure tunnel
• check_circleRunaway-loop monitoring wired to Datadog and surfaced in an ops dashboard
• check_circleConnected to workflow automation and an AI assistant connector
• check_circleOne of multiple MCP servers shipped and running live across engagements

ProblemInventory, driver kits, allocation, picking, receiving, and shipping operated across disconnected Salesforce records and interfaces, with limited shortage visibility and risky production changes.

BuildA complete fulfillment lifecycle connecting Contacts, kits, Supply Requests, warehouse inventory, automatic allocation, shortage replacement, receipts, pick slips, UPS rates, Work Orders, labels, tracking, and exception dashboards — released through a guarded nine-stage deployment.

ResultFulfillment teams can move an order from request to shipment inside one traceable Salesforce workflow, while preserving allocation history and production-only capabilities.

• check_circleWarehouse-level inventory, adjustments, receipts, thresholds, and allocation
• check_circleBulk and individual pick-slip generation
• check_circlePartial-allocation visibility and unavailable-item replacement
• check_circleUPS rates, labels, tracking, carrier, and service level returned to the request
• check_circleCancellation releases inventory without deleting allocation history
• check_circleNine-stage production rollout with focused validation
• check_circle22 historical shipping-label links repaired

ProblemA medical-transport operator routes inbound calls between live queues and an AI voice agent across several business lines — and was flipping that routing by hand, line by line, every day. Easy to forget, easy to leave a line on the AI overnight, and invisible once it had drifted.

BuildAn in-app control page that toggles each phone line between its normal queue and its AI-forward queue three ways: manually, on recurring day-and-time schedules, and with one-off fail-safe triggers that force a line to a known state at a set time — a safety net that fires even when the scheduler is off. Every switch, human or automated, is written to a persistent activity log. The scheduler is event-driven: it only touches the phone system at real transitions, and it never acts on a state it cannot read.

ResultSet-and-forget call routing across every line, with a master switch that defaults OFF, a full audit trail, and the operator still in control — including two live production outages diagnosed and fixed along the way.

• check_circleRecurring day/time schedules plus one-off fail-safe triggers across four business lines
• check_circleEvent-driven scheduler (per-minute cron + edge-function tick) that makes near-zero vendor API calls between transitions
• check_circlePersistent activity log capturing every manual and automatic switch, with source and actor
• check_circleDiagnosed and fixed live outages — vendor rate-limiting and stale-token invalidation — with a shared token cache and automatic re-auth
• check_circleMaster switch defaults OFF; an unreadable line state is never blindly switched
• check_circleConsolidated onto a single backend and retired a stale duplicate to remove split-brain risk

ProblemA solo speech-language pathologist needed to log sessions on a phone or laptop while staying aligned with HIPAA — without an enterprise EHR's cost or overhead.

BuildA private PWA on self-hosted infrastructure the owner controls, with a green CI pipeline and an explicit runbook + auditor gate that must clear before any real patient data flows.

ResultA clinician-grade logging app built to a compliance bar, with real patient data deliberately gated behind an infrastructure sign-off.

• check_circleHIPAA-aligned design on owner-controlled, self-hosted infrastructure
• check_circleGreen CI pipeline; shipped in defined milestones
• check_circleReal patient data gated behind a written runbook + auditor sign-off

ProblemAn event-rental business needed more than a standard storefront. Customers had to explore a large visual inventory, assemble coordinated looks, and request a quote without the experience feeling like retail checkout.

BuildA custom-branded Shopify experience with curated rental collections, shoppable lookbooks, product recommendations, quote-request workflows, hidden public pricing, and operational tooling for managing catalog content, inventory structure, policies, and rental agreements.

ResultConverted a service-heavy event-rental operation into one structured digital workflow — from visual discovery and product selection through quote submission and signed rental documentation.

• check_circle42 rental products organized into 8 customer-facing collections, turning an unstructured inventory into a browsable catalog
• check_circle4 shoppable looks connect visual inspiration directly to quote-ready product selections
• check_circle1 unified quote journey replaces disconnected product inquiries and conventional buy-now checkout
• check_circle6 branded agreements standardize delivery, will-call pickup, and vehicle-use handoffs across two rental brands
• check_circleCatalog publishing, cleanup, collection assignment, and product recommendations converted into repeatable automation

ProblemLending policies live in long documents containing hard eligibility rules, exceptions, risk tolerances, institutional priorities, and mission-level guidance. Underwriters must interpret those policies consistently for every application.

BuildA Salesforce-native workspace where administrators upload and activate credit policies, review extracted policy interpretations, and evaluate applications against the active policy. Results include pass, attention, and outside-policy findings, policy citations, concise guidance, missing evidence, and recommended next actions.

ResultA governed underwriting-assistance experience embedded directly in Salesforce, with policy versions, evidence provenance, human review, and repeatable application analysis.

• check_circlePolicy-document upload and activation workflow deployed in Salesforce
• check_circleApplication analysis available through both a record action and embedded workspace
• check_circleFindings include policy references, missing evidence, and human-readable guidance
• check_circlePermission-controlled access and Salesforce-native record storage
• check_circleActive-policy provenance preserved with every analysis

ProblemCredit memos are not one-size-fits-all. A construction loan requires project budgets, timelines, collateral, leverage, completion risk, and exit analysis. A working-capital facility requires bank activity, cash flow, DSCR, leverage, and repayment-capacity analysis.

BuildA Salesforce-native Credit Memo generator with governed configurations for different loan programs. Users select the program, generate an editable draft from available application and policy evidence, review missing information, refine every section, and generate a formatted PDF stored in Salesforce Files.

ResultA repeatable memo-generation workflow that adapts document structure and evidence requirements to the selected loan program while preserving configuration, policy, and source provenance.

• check_circleWorking Capital and Construction memo configurations deployed
• check_circleProgram-specific document sections and evidence requirements
• check_circleEditable Salesforce Credit Memo records with complete Details pages
• check_circlePDF generation with automatic Salesforce Files storage
• check_circleConfiguration version, policy version, and evidence provenance recorded
• check_circlePermission-controlled administration and user access

ProblemA broker submission normally kicks off days of back-and-forth — pull the file, check it against the credit box, go back for clarification, wait on an underwriter — before anyone can give the broker an answer.

BuildAn Agentforce decision desk that reads the submission and returns a first-pass call on the spot: approve or decline, a confidence score, and the maximum amount it would approve, all grounded in the real underwriting criteria. Change an input — time in business, credit score, monthly revenue — and it re-runs the decision live; the broker can ask follow-up questions and pull up the underwriter analysis behind the number.

ResultSeconds-not-days first-pass decisions in front of the broker — with a hard line: when an input crosses into risk (a low-enough credit score), the agent flags the submission for human review instead of rubber-stamping it.

• check_circleReads a broker submission and returns approve/decline + confidence score + max approvable amount
• check_circleLive re-run when inputs change, with follow-up Q&A and drill-down to the underwriter analysis
• check_circleEvery call grounded in the actual underwriting criteria
• check_circleRisk-crossing submissions flagged for human review, never auto-finalized

ProblemOperations-heavy businesses — dispatch, NEMT, home services — lose revenue to missed and after-hours calls, and the humans who answer burn out on repetitive intake.

BuildA voice-AI product with per-vertical demo agents you can call from the site, a gated agent map that routes each industry to the right configuration, and a marketing site migrated entirely onto the same modern stack.

ResultLive at voitra.ai with callable vertical demos, fully off the old Webflow site and serving every route from one Next.js Worker codebase.

• check_circleLive and callable at voitra.ai
• check_circle100% migrated off Webflow — all routes serve from the Next.js Worker
• check_circlePer-vertical demo agents wired to a gated agent map

ProblemSolo operators run their business across a dozen tools with no system of record. The hard part isn't building one for yourself — it's turning that personal tool into a product other operators can run inside, without their data ever touching each other's.

BuildRe-architected the single-user app into a multi-tenant platform with row-level tenant isolation, then onboarded the first external tenant onto its own gated instance and subdomain. Each client gets a modular set of control surfaces — telephony queue routing, property/HOA data, team-chat operations, custom evidence trackers — switched on per account. Added per-tenant usage analytics, an owner-only cross-tenant admin, one-click self-serve connections (MCP-to-AI and Gmail), and an AI knowledge graph that reads a tenant's own notes and records and maps how everything connects.

ResultOne codebase now serves multiple isolated operator instances live, each on its own domain, with per-client capabilities and self-serve setup — a repeatable product motion, not just a personal tool.

• check_circleMulti-tenant with row-level data isolation — no tenant can see another's data
• check_circleFirst external tenant live on its own gated subdomain
• check_circleModular per-client control surfaces, toggled per account
• check_circlePer-tenant usage analytics + owner-only cross-tenant admin
• check_circleSelf-serve MCP-to-AI and Gmail connections, per tenant
• check_circleAI knowledge graph that extracts entities + relationships from a tenant's own data

ProblemFive families coordinate a school run over group chat. Nobody knows who is driving, where the car is, or whether to send a child outside yet. A wrong guess means a child waiting alone on a pavement.

BuildA carpool app on Cloudflare Workers and D1, delivered as a website and as an Android app on Play internal testing. Alerts go over Web Push and Firebase Cloud Messaging together, so one person is reached whether they use the site on a laptop or the app on a phone. Pickups complete themselves when the car reaches the house, and each driver sees the route in the order they actually drive it.

ResultIn use every school morning by the families it was built for, with the route, the alerts and the data model all shaped by what went wrong in real runs.

• check_circleLive at carpoolcircle.com, driven every school morning
• check_circleAndroid app on Google Play internal testing; iOS built and verified on simulator, pending Apple enrollment
• check_circleAlerts delivered over both Web Push (RFC 8291) and FCM, reaching Apple and Android devices from one send
• check_circleBackground location on Android via a foreground service, which a browser cannot do
• check_circlePickups complete on arrival, with a radius derived from the route's own geometry
• check_circleMigrated live off a hosted builder onto owned infrastructure with two independent restore paths
• check_circlePush built against RFC 8291 directly, after finding that two edge libraries ship a pre-standard scheme Apple rejects
• check_circleFCM v1 spoken over a WebCrypto-signed JWT rather than pulling a server SDK into a Worker

ProblemAugmentative & alternative communication (AAC) tools for nonverbal children are often expensive, clunky, and locked to a device.

BuildA progressive web app with configurable board sets and per-child profiles, installable on any device, backed by Supabase.

ResultA live, installable AAC board that works across devices at no hardware lock-in.

• check_circleLive PWA, installable on any device
• check_circleConfigurable board sets + per-child profiles on a real backend

ProblemOperators want a point of view they can act on, plus a fast way to see where their own operation is leaking — without booking a consultant first.

BuildA publishing + intelligence platform: an automated morning-brief Worker that writes to a Supabase CMS, an 'Ask Aventary' retrieval assistant over a knowledge base, and interactive diagnostic tools (lead-to-opp, operating diagnostic) — all on the site you're reading.

ResultA living asset that generates content on a schedule and lets prospects self-diagnose before the first call.

• check_circleAutomated morning-brief Worker publishing to a live CMS
• check_circle'Ask Aventary' retrieval assistant grounded in a curated knowledge base
• check_circleSelf-serve diagnostic tools live on the site