Brandon Ifill
AI Engineer — Agents · Evals · RAG · MCP | Full-Stack (TypeScript / Python)
Self-taught AI engineer who ships. Built and deployed an AI agent platform (30+ agents, multi-provider failover, RAG), a multi-model citation-verification engine that passed zero bad citations on an adversarial test set, a five-module MCP computer-use stack, and an AI IDE on a VS Code fork — 15+ applications across 6 industry verticals, all as a one-person full-stack team. Evals and honest measurement built in from the start. Ships daily.
Featured Projects
Verity
Legal-Citation Verification Engine | 2026
- •Fail-safe verification for AI-generated legal citations: a diverse 5-model panel (GPT-4o, Gemini, Llama, Mistral, DeepSeek) cross-checked against CourtListener ground truth, with a PASS-only-on-agreement rule
- •On a 27-case adversarial test set — including the real Mata v. Avianca fabrications — the combined system silently passed 0 bad citations, while single models alone endorsed up to 10 fabricated cases
- •Documented the core eval findings: narrow model panels share hallucinations; model diversity plus external ground truth is the only safe design
- •Reproducible eval scripts with published test sets and written results
CTRL-A Workspace
AI Agent Platform | 2026–Present
- •Built a proprietary AI platform from scratch — 30+ specialized agents built for 6 business verticals (CPA, healthcare, legal, e-commerce, auto, general)
- •Agents use tool-calling, vector memory (RAG), sentiment detection, and automatic failover across multiple AI providers (Anthropic, OpenAI, local models)
- •Real-time chat via WebSocket, conversation history compression, knowledge base ingestion with document chunking, and agent-to-agent delegation
- •Fleet management system: built to deploy CTRL-A to multiple businesses with central monitoring, heartbeat check-ins, version reporting, and remote updates without downtime
- •Multi-tenancy isolation hardened through a documented audit-and-fix ledger
- •Express/TypeScript backend, Turso (libSQL) database, Docker, rate limiting, Zod validation, audit logging — deployed on Render with continuous delivery
CTRL-A Computer Use
Agent Computer-Control Stack — 5 MCP Servers | 2026
- •A layered stack that gives any AI agent control of a real computer: visual (screenshot + mouse/keyboard), native OS accessibility-tree element targeting, file/SQLite state reading, and app plugin-API bridges
- •Includes a Ghidra pipeline that reverse-engineers a binary and generates Frida hooks plus an MCP server for its internals — integration even when the target app has no API
- •License gating on every module: Ed25519-signed session JWTs with expiry, machine binding, and a revocation denylist re-checked on every tool call
- •Forked VS Code OSS into an AI coding environment with the agent integrated into the editor itself — not a bolted-on sidebar
- •Gateway panel for real-time monitoring of agents, skills, tools, and sessions; full build pipeline (gulp, CodeQL, third-party license management)
- •Upload your docs → AI builds a learning path with interactive labs, AI-graded exercises with partial credit, and a grounded tutor that answers only from the supplied sources
- •Next.js 16, Stripe billing, NextAuth, Turso persistence, Telegram channel integration — with type-check/test/build gates on every deploy and a Playwright e2e suite
- •Production management built from 7 years inside the film industry: call sheets, crew management, scheduling, budgets, media
- •Systematic security-hardening campaign: IDOR remediation across 8+ surfaces, cast PII protection, RBAC, schema paydown — tracked in per-issue worktrees with audit reports
- •Next.js with Firebase, deployed
- •PWA with Plaid bank integration, CTRL-A conversational AI coaching, and 30/60/90-day cash-flow forecasting with "What If" scenario modeling
- •Clean-room design respecting client patent (US 10,210,583 B2)
Project Orange Blossom
Food Truck Ordering Platform | 2025–2026
- •Full-stack mobile ordering with a 70+ table schema, tRPC API, Stripe payments, Ably real-time order tracking, Leaflet maps, PWA
- •Migrated Supabase → Turso with a custom Supabase-compatible query adapter
- •Community business directory with e-commerce, memberships, payments, messaging, bookings, and an AI shopping assistant — installable PWA
Technical Skills
AI / LLM Engineering
Anthropic & OpenAI APIs, OpenRouter, multi-provider failover, agents & tool-calling, MCP server development, RAG (embeddings, vector search, citation-grounded answers), evals (LLM-as-judge, multi-model panels, ground-truth verification), prompt engineering, computer-use automation
Languages
TypeScript, JavaScript, Python, SQL, HTML/CSS
Frontend
React 19, Next.js 14–16, Tailwind CSS, shadcn/ui, Framer Motion, Recharts, PWA
Backend
Node.js, Express, tRPC, REST APIs, WebSocket, Zod validation, rate limiting, audit logging
Databases
Turso (libSQL), SQLite, PostgreSQL (Supabase), Drizzle ORM
Integrations
Plaid, Stripe (subscriptions, webhooks, Connect), Ably, Socket.io, Telegram
Auth & Security
Firebase Auth, NextAuth.js, JWT (jose), Ed25519 signing, bcrypt, RBAC, IDOR remediation, PII protection
DevOps & QA
Docker, Render, Vercel, Firebase Hosting, DigitalOcean, GitHub CI, Playwright e2e, type-check/test/build deploy gates
Professional Experience
Founder & AI Engineer — Sage AAA
Jan 2025–Present
- • Founded an independent studio building AI-enabled platforms and internal tools — sole architect and engineer
- • Shipped 15+ applications across finance, film, food service, legal, and community verticals
- • Built CTRL-A (agent platform), Verity (citation verification), DevClaw (AI IDE), and a five-module MCP computer-use stack
- • Own the full lifecycle: discovery, architecture, build, deploy, monitoring, and security review
Film Industry Professional — Atlanta, GA
2017–Present
- • Seven-plus years working inside film and television production
- • Translated first-hand workflow gaps into SlateFusion and other production tools
CDL Truck Driver — Various
2003–2024
- • Two decades of over-the-road logistics: scheduling, routing, DOT compliance
- • Independent problem-solving and operational discipline, applied directly to software engineering
- • Self-studied programming during off-hours, building first projects in 2024
Education & Development
- • Self-taught full-stack & AI development (2024–present)
- • TOR 90 Career Accelerator Program participant (2026)
- • Continuous focus: agent systems, eval methodology, MCP, RAG, cloud architecture, security hardening
Don't take my word for it — use it.
This site, the client portal, the AI chat, the proposal system — everything on this resume is running in production.