About MAX EV Digital

Problem: Needed a unified AI-native business OS to consolidate and replace a full legacy SaaS ecosystem — CRM, helpdesk, e-signature, scheduling, email, and invoicing — across multiple client verticals.

• ›50+ live AI endpoints running MCP protocol, multi-model routing (Claude + GPT-4o), HITL guardrails, and Zod-validated output schemas
• ›Per-request observability: model ID, prompt version, latency, raw output, and correction signals logged on every AI call
• ›BullMQ + Redis drip engine, Stripe billing + webhooks, Google Calendar OAuth2, custom 3-panel IMAP email client
• ›Not a demo — enterprise-grade system running against live operational data daily

Problem: Senior engineers managing active job searches waste hours manually hunting boards while losing track of applications, follow-ups, and concurrent income. No tool handled discovery, scoring, HITL authorization, and pipeline tracking in one autonomous loop.

• ›IMAP IDLE inbox monitoring with real-time classification; BullMQ parallel worker processing (9 workers running 24/7)
• ›Claude Haiku scores every opportunity 0–100 at $0.003/score; Sonnet drafts cover letters at score 80+ and delivers PDF via Telegram
• ›35+ company ATS APIs polled daily (Greenhouse, Lever, Ashby) — Anthropic, Stripe, Figma, Scale AI and more
• ›2,100+ opportunities processed; headless Chromium autofill pipeline; Telegram HITL loop for one-tap Apply/Skip/Later

Problem: Regional field services contractor was paying $2,000+/month in fragmented SaaS across estimate, dispatch, CRM, and outreach tools — with no connection between them and no outreach automation.

• ›Full ops admin panel: estimate wizard, job scheduling, crew dispatch, contract tracking, billing, customer portal
• ›Twilio IVR with AI-scripted call routing + Sinch SMS batch campaigns triggered by TCAD property data pipeline (1.4M+ parcels, 3 counties)
• ›ElevenLabs TTS voice agent + Storm Alert outreach integration; lead intake to CRM to dispatch loop
• ›Replaced $2,000+/month SaaS stack; delivered outreach automation in under 2 weeks

Problem: Financial wellness coaching at scale required a RAG-based system that grounds every response in structured financial data — with fiduciary guardrails that prevent hallucination and ensure source-cited answers.

• ›FastAPI backend with pgvector semantic retrieval — document chunking, embedding pipeline, and hybrid retrieval
• ›Claude API with structured context injection; prompt evaluation layer for response quality benchmarking
• ›Fiduciary guardrail system: confidence thresholding, source citation enforcement, no hallucination surface
• ›Built to demonstrate RAG integration in a compliance-sensitive environment with grounded, auditable outputs

Problem: Demonstrate end-to-end AI-native engineering for enterprise loan review — connecting AI capabilities to live business systems (Snowflake, Salesforce, Gmail, Slack, Zoom) with auditable outputs and HITL gates before any consequential action.

• ›FastAPI backend with MCP tool registry (5 tools: Snowflake, Salesforce, Gmail, Slack, Zoom) — called via call_tool(), never directly
• ›Multi-model routing: Haiku for fast classification, Sonnet for judgment and structured decisions
• ›HITL confirmation gates for all external actions (email send, Slack post, Zoom create) with Pydantic-validated outputs
• ›Full mock mode (MOCK_MODE=true) for demo — same agent logic, routing, and validation against realistic mock data

Problem: EdTech platform needed a real-time AI coaching system with a custom voice persona, live broadcast infrastructure, and Stripe-gated membership for craps strategy instruction.

• ›Real-time strategy simulation engine with AI coaching loop; marathon session mode for extended play analysis
• ›ElevenLabs TTS custom voice persona (Pulse) — never auto-regenerated, triggered only on explicit user action
• ›Live broadcast infrastructure for StreamYard-based instructional delivery; GA4 analytics integration
• ›Stripe-gated membership; PM2 on VPS; ATS overlay ($25 Small / $25 Tall / $10 All bet sizing)

Problem: The gaming industry had no dedicated career and news platform — job seekers scattered across generic boards, properties with no niche channel to reach qualified candidates, and no editorial voice covering casino operations across 50+ US gaming markets.

• ›Combined job board with employer profiles, applicant career resources, and automated industry news feed
• ›PostgreSQL backend with full CRUD for job listings, employer branding pages, and editorial content
• ›Dual revenue model: job postings + employer profile subscriptions; success fee structure for placements
• ›Forked from DFW Daily codebase — multi-tenant platform architecture shared across two verticals

Problem: DFW businesses and residents lacked a local-first news and business directory platform with AI-generated editorial briefs.

• ›News aggregator + business directory with AI-generated DFW-specific briefs (Claude)
• ›Category system across 25 industry verticals; admin panel for editorial management
• ›Standalone deploy script (deploy.sh) for Next.js standalone mode with static asset copy
• ›45,000-business DFW database backend for targeted local outreach and directory population

Problem: Regional home services company needed an end-to-end platform — from a public-facing marketing site with self-service estimate tool through to a full admin panel for job pipeline and status tracking.

• ›Public Next.js marketing site with multi-step estimate tool and smart lead capture form
• ›CRM routing logic: leads auto-assigned based on service type and geographic zone
• ›Admin panel for job pipeline tracking, job status updates, and client communication history
• ›Self-contained — no third-party SaaS dependency for CRM, scheduling, or customer-facing intake