Co-Founder & CTO at GVVA, an applied-AI engineering firm. I'm a hands-on engineer first: I design, build, and operate production AI systems end-to-end — the agents, the cloud infrastructure underneath them, and the compliance and reliability engineering that lets them touch real customers.
I'm not a demo-builder. My flagship engagement is a production, multi-agent AI platform that runs the entire lead-intake operation of a U.S. personal-injury law firm — from the first website chat to a signed retainer to case-work documents — live since 2025, co-built and operated with Masego Letsoko (@SegoML).
That means we've had to be good at the whole stack:
- AI agents in production — Amazon Lex V2 conversational intake (bilingual EN/ES), an Amazon Bedrock (Claude) decision engine that triages thousands of CRM records, and a 10-stage AI paralegal pipeline that generates and sends legal correspondence.
- Serverless AWS at depth — 25 Lambda functions, API Gateway + Cognito (JWT, MFA), 8 DynamoDB tables, EventBridge schedules, S3/CloudFront, Secrets Manager, Transcribe, SES — deployed with AWS SAM.
- Telephony & compliance engineering — a TCPA-aware outbound voice engine on Twilio Programmable Voice: answering-machine detection, deterministic DTMF-only IVR, BYOC SIP trunking, consent gating, and a full audit trail (call recording → transcription → archived PDF call summary) designed for legal defensibility.
- Marketing attribution — server-side Meta Conversions API with pixel/CAPI deduplication, hashed PII matching, and a multi-source lead classifier across Google, YouTube, Meta, and TikTok traffic.
- Frontend product — a real-time operations dashboard in React 19 + TypeScript + Vite + Tailwind + TanStack Query, used daily by the firm's staff.
- Reliability & forensics — rate-limit engineering against hard API quotas, fail-safe AI design (an AI error must never silently drop a lead), versioned backup regimes, and root-cause investigations that separate code failures from campaign-config failures with evidence.
| Repo | What it shows |
|---|---|
| legal-intake-ai-platform | Case study of the full three-agent platform: architecture, data model, and the engineering decisions behind it |
| gvva-multiverse | The AI operating system GVVA runs itself on: Nectar (culturally-adaptive intake concierge), Guava Pulp (25-Lambda Claude-powered ops engine), SEED (founder dashboard) |
| tcpa-compliant-voice-engine | Deep dive on the outbound voice system: why we migrated off a voice-AI vendor to raw Twilio, and how the compliance audit trail works |
| serverless-ad-attribution | Server-side Meta CAPI + multi-domain pixel routing + lead-source classification, and a marketing-forensics post-mortem methodology |
These are case studies of live client systems. Client-identifying details are anonymized and code samples are sanitized excerpts — the production codebases are private, as they should be. I'm happy to walk through the real systems in a technical interview.
- Production is the bar. A system isn't done when the demo works; it's done when it survives real users, real edge cases, and real regulators.
- Fail safe, not silent. When the AI layer fails, the system must degrade to a human decision — never drop a customer on the floor.
- Verify, don't trust. Every deploy is verified against live behavior; every incident gets a root cause, not a guess.
- Compliance is a feature. TCPA, consent, and auditability were designed in, not bolted on.
Python TypeScript React AWS Lambda AWS SAM DynamoDB API Gateway Cognito Amazon Bedrock Amazon Lex Amazon Connect Amazon Transcribe SES EventBridge S3/CloudFront Twilio Voice SIP/BYOC ElevenLabs Meta CAPI GA4 Vite Tailwind TanStack Query