Client results first: case studies with numbers from the project files, client work anonymized and my own products named. Then more of my builds, client work included, sorted by type: Claude agents, AI agents and voice, CRM and outreach, marketing and content, documents and operations, and full-stack apps. Client names are left out on purpose; the numbers are not.
The client builds I would show you first, with the numbers.
A custom CRM on Lovable, Supabase and n8n. A Claude agent drafts outreach into Outlook and has no way to send it. The client clicks send.
An n8n, Otter and OpenAI pipeline that grades every recorded consultation, files a PDF report and emails a manager only when a grade is yellow or red.
Each one written up in full, with numbers from the project files.
AI stock research where every proposal waits for a human, and a site published daily from June 16 to October 8, 2026. A research tool, not financial advice.
A to-scale 3D copy of IBA Karachi's main campus in the browser, built by an alumnus. 903 people signed up in its first ten days, July 2026, and 97% walked in.
A budget-gated intelligence agent, a monitoring agent that cannot invent a fact, a document librarian that never deletes, and a daily action list.
n8n, Postgres, Qdrant, private search and a filings API on one free Oracle ARM server (4 cores and 24 GB, the 2025 allowance), with HTTPS and a runbook.
Four listing templates with layout locked in code, an editable mode that matches the PDF to the pixel, and a Claude Code handover for a non-technical agent.
Builds where Claude or Claude Code does the thinking: screening, research, triage and writing.
Screens a list of companies against the client's own written rules with parallel Claude agents, an independent review and the client's validator, one published checkpoint at a time.
Scored every company in a target list with the client's own fit prompt on Claude, then sent each Strong verdict to a skeptical second judge before it counted.
Three scheduled Claude tasks for a solo real-estate agent: a weekly market pulse, four social posts a week and three past-client email templates a month, each emailed only to her.
Reads and classifies every incoming email with Claude and surfaces a prioritized view, with escalation rules the client controls.
Agents and assistants that answer, research and plan, by chat, email, voice or over your own data.
Calls real estate leads, captures budget, timeline and buyer profile, and logs an AI call summary to the CRM sheet.
Ingests Slack threads and Notion pages into a vector store and answers the team's questions from them.
Three voice agents I built: a support agent, an appointment booker and my own studio assistant. Live calls are paused for now.
One agent answers and qualifies leads across WhatsApp, Instagram and a website chat, and routes each conversation to the right place.
Turns plain-English questions into SQL and runs them against Supabase Postgres, so people who do not write SQL can pull live data.
Reads support emails, classifies intent, pulls the Shopify order and drafts replies grounded in policy and FAQ data.
A browser agent on Airtop that runs searches on sites like LinkedIn, extracts the results and writes them to Google Sheets.
Systems that capture leads, keep CRM data clean and complete, and draft or send the follow-up.
The deal-source database inside the client's CRM: investment firms and their portfolio companies, with filters, bulk assignment, notes and exports in his report template.
Fills blank contact fields from email signatures, web search and firm bio pages, only above a confidence gate, with an independent verifier and one-click undo.
Admin tools inside the client's CRM to normalise values, merge duplicate contacts safely and bulk-import CSV files with a preview and a diff, all audit-logged.
GoHighLevel survey leads pass ZIP gating, cleanup and a duplicate check in Zapier and Google Sheets, then post to LeadConduit.
Three n8n milestones for a US GoHighLevel agency: nightly call attribution, a loyalty-member import that can safely run twice, and a chatbot warehouse.
The n8n stack behind this site: lead capture, a timed follow-up sequence that stops when someone books, the chat assistant, the demo backends and a daily health check.
Drafts personalized emails for human review from CRM contacts due on 30, 60 and 90+ day cycles.
Pipelines for marketing work: SEO and topic research, drafts, posts, newsletters, flyers and video.
A monthly newsletter engine for a solo real-estate agent: Claude writes the issue, and code owns the layout, legal lines and fact checks, in three brand-locked styles.
Turns the week's market numbers into a branded 1080 by 1350 flyer, and refuses to render if text clips, a source is missing or a number is not in the saved report.
Turns a paid student form submission into a post on the right college Instagram page through the Meta Graph API.
One English script becomes localized UGC-style video ads on Google Veo 3, run from a Google Sheet and delivered to Google Drive.
Submit a URL and get an AI-written SEO audit. It doubles as a lead magnet that captures every request.
Creates and publishes blogs, LinkedIn posts, faceless videos and images, all controlled from Slack.
Pulls trending ideas from chosen subreddits, processes them with OpenAI and files the results in Airtable.
Turns a one-line idea into a detailed prompt, renders it with Sora 2 as an async job and emails the finished video link.
Writes scripts with OpenAI, renders avatar videos in HeyGen and publishes them through Blotato.
Fetches posts on chosen topics from the X API, summarizes them with an AI agent and emails a morning digest.
Keyword in, researched draft out: Perplexity research, a Claude draft and an OpenAI critique, saved to Google Docs and Gmail for a person to publish.
Enter a keyword and a location to pull DataForSEO keyword data in one run, filed in a Google Sheet and summarized for the front end.
Systems that read documents, build reports, check and move data, and keep the back office running.
Filed a client's cluttered cloud drive into a structure he signed off, in reviewed batches, with every move journaled and nothing deleted.
Corrected the client's company database under one rule, a blank cell beats a wrong one, measuring the error of every field and rolling back any field that crossed 15 percent.
An independent audit of the AI screening output, built when the client asked for one: ten audit families, with every defect logged and classified by root cause.
Read each investment firm's website for its portfolio companies, queued them for review without duplicates, and promoted the approved ones in one batch a single call can undo.
Rebuilds the client's nine-tab Excel workbook from every checkpoint of the screening run, so the report always matches the latest validated data.
One click in the results app builds a 16:9 briefing deck from a frozen snapshot of the data, with text measured to fit every slide.
Drop a file into a chosen OneDrive folder and a clean Markdown copy comes back, typically in 30 to 60 seconds: the document pipeline from the case study "Four AI systems for one consultant".
A button on the CRM's Reports page that exports three groups of opportunities as one styled Excel workbook, reconciled against an earlier verified file.
Weekly Excel reports per lead, a Monday master report and a one-page monthly summary, emailed to the client only, plus a round trip that reads his edits back.
Writes reference-backed reports from notes, uploaded documents and fixed writing guidelines, then fact-checks its own draft.
Sends rent reminders, overdue notices and follow-ups on business days only, with holiday logic and CRM sync.
Follows up on overdue and pending Microsoft To Do tasks with personalized reminders by Gmail or Outlook.
Extracts invoice and receipt data with OCR and AI, then routes approvals automatically.
Products with their own interface, on the web or the desktop.
A web app where invited investment reviewers browse the screening results, with filters, search, row drawers and Excel exports.
Lovable front end, Supabase sign-in with row-level security, private chat history and hands-free voice input through n8n and Whisper, for program members.
Open-source Windows dictation app: local Whisper transcription with optional LLM cleanup, no subscription.
Recruiter and client portal on React and Supabase: talent shortlisting and structured client feedback over the Liveforce API.
Booking-style travel app with AI trip recommendations and itinerary generation, built on Lovable and Supabase.