I design and build AI systems. Not tools, not demos — full systems. The kind that replace operational bottlenecks, run 24/7, and connect every part of a business from lead generation to closing to reporting.
My background is engineering, but my thinking is systems-level. Before I write a line of code, I map the business process, find where the biggest leverage is, and design around that. Then I build it and ship it.
I've delivered 100+ production AI systems — voice agents processing hundreds of leads monthly, multi-agent LinkedIn engines that scaled a network while maintaining zero violations, WhatsApp assistants resolving 93% of queries without human input. Every one of them runs live, not as a proof of concept.
What I'm working on now:
- AI infrastructure at Trilles AI
- Voice AI pipelines for sales and qualification
- Multi-agent systems designed to operate without supervision
| What | How It Works in Practice | |
|---|---|---|
| 🧠 | AI Agents & Multi-Agent Systems | Agents built on CrewAI, LangGraph, and AutoGen that plan, delegate, and execute across real business workflows |
| 🎙️ | Voice AI Systems | Inbound and outbound AI callers via Vapi, RetellAI, LiveKit, and Twilio that qualify leads and move deals through pipelines |
| ⚡ | Workflow Automation | n8n, Make, and Zapier systems that wire together LLMs, CRMs, APIs, and databases to run without babysitting |
| 📊 | Business Intelligence | Pipelines that score leads, generate reports, and surface decisions rather than just data |
| 🌐 | Full-Stack AI Products | React and Next.js frontends with AI backends, real-time dashboards, and CRM integrations |
| 🎬 | Generative Media Pipelines | Automated systems that go from product brief to UGC video to ad creative at scale |
Languages
AI Frameworks & LLMs
Voice & Communications
Automation
Frontend & Backend
Data & Infrastructure
🔗 LinkedIn Lead Generation System — 1,500 to 3,000+ Connections, Zero Violations
The gap: B2B teams hit a ceiling on manual LinkedIn prospecting while automation attempts risk account bans.
The system: Three interconnected workflows. Scrapes engagement on competitor posts. Filters to decision-makers only. Deduplicates globally across all campaigns. Verifies outreach history via HeyReach before any message goes out. Sentiment analysis routes interested replies to Calendly instantly. Custom Next.js dashboard tracks campaigns in real time.
What it delivered: 1,500 to 3,000+ connections. Thousands of leads processed weekly. Zero LinkedIn violations. Response time from hours to seconds.
n8n Apify HeyReach Next.js TypeScript ChatGPT PostgreSQL
🎙️ AI Voice Qualification System — 750+ Leads Processed Per Month, Live
The gap: Sales teams waste hours manually calling dormant mortgage leads with inconsistent follow-up.
The system: Pulls dormant leads from CRM automatically. Runs outbound and inbound call flows. Qualifies on property, mortgage, and credit criteria. Auto-advances pipeline stages based on call outcome. Double-attempt logic, voicemail detection, and delayed follow-up built in.
What it delivered: 750+ leads processed monthly. Full call logging for compliance. Manual cold-calling eliminated.
n8n RetellAI Vapi GoHighLevel PostgreSQL Zapier
📱 WhatsApp AI Assistant — 93% of Queries Handled Without Human Input
The gap: High-volume WhatsApp support caused slow replies, missed orders, and inconsistent follow-up across languages.
The system: Multilingual WhatsApp Business assistant. Handles product queries, creates orders and bills, detects crop disease from photos, runs admin broadcasts, and books meetings. Covers routine queries end-to-end.
What it delivered: 93% of queries auto-resolved. 70% reduction in manual replies. 24/7 sales and support coverage.
n8n WhatsApp Cloud API Python Gemini ChatGPT Groq Google Sheets
🎬 UGC Video Ad Pipeline — Product Photo to Finished Ad in Minutes
The gap: UGC-style video ads convert well but require creators, editing, and budget. Impossible to scale.
The system: Two-mode pipeline. Mode 1 analyses a product image, generates a UGC scene, and creates an 8-second video. Mode 2 animates between two keyframes with AI-generated transition prompts. Polling and retry logic for reliability at scale.
What it delivered: Product brief to campaign-ready creative in minutes. Creative testing at scale without a production team.
n8n GPT-4o Claude Fal.ai Google Drive
📈 Stock Market Intelligence System — Full Research in Seconds
The gap: Traders switch between multiple tools for technicals, fundamentals, and news. Slows decisions and loses context.
The system: Single webhook that fetches multi-timeframe candles, fundamentals, and recent news in parallel. Generates chart images. Delivers structured short-term and long-term recommendations through a dual-LLM pipeline.
What it delivered: Full research compressed to seconds. Consistent risk framing on every output.
n8n TwelveData NewsAPI Chart-IMG GPT-4o Mini Claude
📣 LinkedIn Content System — Research, Write, Approve, Post
The gap: Good LinkedIn content takes research, writing, and design time. Most AI-generated posts read like AI.
The system: Phase 1 scrapes top-performing posts by topic. Phase 2 runs a research agent for fresh stats and angles, then a writer agent generates posts and images. Phase 3 is a Sheets-based approval loop where the human adds feedback, and the system revises and regenerates.
What it delivered: Research-backed posts in the brand's voice, with human oversight, running consistently.
n8n Apify Claude Gemini GPT-4 Mini Google Sheets
Happy to talk about system design, AI architecture, or what it takes to actually ship something that runs.
