AI automation and agents that do real work inside your business
Burhan Khatri builds AI automation for businesses: agents that reply to leads and book appointments, pipelines that extract facts from documents with citations, and LLM-powered tools wired into the CRM, inbox and calendar a company already uses. Projects use Claude, OpenAI and open models through Python, FastAPI, n8n and TypeScript.
The useful version of AI in a small business is rarely a chatbot. It is a lead that gets a reply in seconds instead of hours, a lease that is abstracted with every fact traceable to the source page, or a 3,000-row bank statement turned into a plan in about a minute.
Each project starts with the workflow, not the model: what comes in, what decision is made, what goes out, and how a human checks it. The model is chosen afterwards and swapped when a better one appears.
What this covers
- AI dispatchers and lead follow-up: reply, qualify, book, and update the CRM (n8n, GoHighLevel, Claude)
- Document extraction with citation checks for contracts, leases and statements
- Multi-agent pipelines for analysis and reporting (CrewAI, Claude, FastAPI)
- Lead generation and enrichment pipelines pushed into outreach tools
- Voice and transcription tools: dictation to structured reports with Deepgram and Claude
- Native macOS agents in Swift that act on the screen, with a Python sidecar for the model
Projects of this kind
- AI Dispatcher for Plumbing & HVAC24/7 dispatcher that replies to every lead in seconds, qualifies the job, books appointments and sends SMS follow-ups
- Precision GTM EngineLead-generation pipeline for a commercial energy broker: discovery, verification, CRM dedupe, enrichment and outreach
- Lease IntelligenceCitation-verified lease abstraction for commercial real estate, with amendments reconciled against the base lease
- BaqiAIMulti-agent platform that turns a bank statement into an investable surplus and a stock portfolio; winner, JS Bank PROCOM '26 AI in BankingVisit BaqiAI
- PathReportVoice-to-report tool for pathologists using Deepgram and ClaudeVisit PathReport
- OrangemacOS voice agent that executes real actions through the Accessibility APIVisit Orange
Frequently asked questions
What can AI automation actually do for a small business?
- The reliable wins are repetitive, text-heavy tasks with a clear rule: answering and qualifying inbound leads, booking appointments, pulling facts out of documents, drafting follow-ups, and keeping a CRM up to date. If a task needs judgment every time, the automation drafts and a person approves.
Will it replace my staff?
- No. The projects above remove waiting and copy-paste work so the same team handles more. A dispatcher that answers at 2 a.m. does not replace the person who runs the job the next morning.
Which AI models do you use?
- Claude, OpenAI models, Groq-hosted open models and on-device models, chosen per task for accuracy, cost and privacy. The integration is written so the model can be swapped without rebuilding the workflow.
Is my data safe?
- Data stays in your own accounts (CRM, database, cloud) and only the fields a step needs are sent to a model. Where privacy matters most, processing runs on-device or on your own infrastructure, as in the macOS tools built for voice and chat data.
How long does an AI automation project take?
- A single workflow, such as lead follow-up into a CRM, typically ships in one to three weeks. Larger pipelines with document extraction or multiple agents are delivered in milestones with a working version early.
How do we start?
- Describe the workflow and the tools it touches on the contact page. You get a short written plan with scope, timeline and price before anything is built.