Things I've built.

Multi-agent AI automation platform combining RAG retrieval, a custom MCP server, RBAC, and human-in-the-loop approval, integrated with CRM, project-management, communications, and payments systems.

5-agent LangGraph pipeline (Research → Sales → CRM → Communication → Reporting) with human-in-the-loop approval, exposed via REST API, n8n webhook, and MCP.

A custom MCP server exposing search, database, CRM, email, calendar, and analytics tools to any AI agent — authenticated, tested, and Docker-ready.

An AI sales agent using LangGraph, GoHighLevel CRM, and n8n to qualify leads, score them, and follow up — with every action gated behind human approval.

Multi-agent architecture with separate Gmail and Calendar sub-agents — handles email (send/reply/search/delete/label) and calendar (create/reschedule/cancel/check availability) from text or voice commands, with contact lookup and cross-turn memory.

Natural-language agent that looks up and verifies a recipient in Google Contacts, drafts a context-aware email, and sends it via Gmail — holding memory across turns to handle multi-step requests.

Two linked pipelines: a knowledge-base pipeline that syncs Google Drive documents into Pinecone as vector embeddings, and a chatbot pipeline running semantic search with memory-enabled RAG responses. Includes an Apify variant for scraping live site content.

Live, publicly deployed chatbot on aismartgalaxy.com that scrapes real site content on each query and answers only from that data, with session memory and a clean fallback for unknown answers.

Conversational AI that answers service questions, qualifies leads, and detects booking/complaint intent via structured JSON output, with auto-tagging and instant human handoff for complex conversations.

Instant SMS, email, and task creation within 60 seconds of a form submission, with a 5-minute manager-escalation check and 24-hour nurture fallback for unresponsive leads.

A missed inbound call auto-triggers a business-hours-aware SMS with a booking link, callback task creation, internal rep notification, and 2hr/24hr follow-up escalation.

A fully pre-configured template sub-account bundling pipelines, custom fields, calendars, and 4 core automations into a one-click deploy — cuts client onboarding from ~5 days of manual setup to ~15 minutes.

Captures inbound leads via Google Forms, syncs to Apollo as CRM contacts, and runs parallel enrichment (structured lookup + Apify LinkedIn scraping) synthesized by GPT into a ready-to-use research summary.

Automates the full candidate lifecycle — captures applications via Gmail, extracts job titles with AI, routes candidates through stage detection, and fires the right action per stage: rejection, screening invite, Slack alert, or interview scheduling.

Watches new applications, converts resumes from PDF to text, scores each candidate against a fixed weighted rubric, and auto-routes to Rejected or Interview-Invite — with an optional manual-override scenario for hiring managers.

Auto-extracts merchant, line items, taxes, and totals from invoice/receipt PDFs and images uploaded via Telegram, structures the data to JSON, logs it to Sheets, and sends a confirmation.

A research agent grounds content in real-time data, a writing agent drafts the post, a design agent creates a matching visual, and a headline agent crafts the hook — publishing straight to LinkedIn with zero manual steps.

Routes incoming WhatsApp messages by type, processes text, images, voice notes, video, and documents, then uses an AI sales agent with product and order tools to prepare the right reply.

Receives Messenger conversations, checks routing and pause rules, handles supported message types, and connects an AI agent to product, order, and customer-service tools.

A reusable lead-generation workflow that accepts a search configuration, gathers local business data, enriches and normalizes records, checks for duplicates, and writes the results to a sheet.

Collects offer data from configured sources and screenshots, extracts and normalizes details, checks existing records, then stores new results and sends notifications.

Checks for new WordPress articles, prepares a Facebook-ready image and post, publishes the content, records completion, and routes failures to an error handler.

A SaaS product concept and interface for a business AI agent, including a unified multi-channel inbox, customer context, agent controls, and operational dashboards.

A seller-operations dashboard concept showing product listings, inventory, supplier costs, pricing, profit, synchronization status, and operational alerts in one view.

A voice-agent concept for handling inbound qualification and calendar scheduling with conversational turn-taking and escalation when a person needs to take over.