I build production-grade automation — email, intake, and document workflows — that gives law firms and service businesses their hours back. Accurate, secure, and built for real data from day one.
Top Rated on Upwork — 20 clients, 27 contracts
n8n, GPT-4o, Claude, AWS Bedrock, Microsoft Graph, Clio, LangGraph
What I build
Three practices, one discipline: take the repetitive, high-volume work off a team's plate without ever putting accuracy or client confidentiality at risk.
Context-aware drafting and routing that turns hours of repetitive email into minutes of review. Connected to your inbox and CRM, running 24/7.
Assistants trained on your private data and agents that take action — update the CRM, trigger workflows, reason across steps. Accurate retrieval, minimal hallucination.
retrieve → reason → act
Automated intake, document generation, and demand-letter pipelines — secure, structured, and built for the volume and accuracy legal work demands.
Selected work
Production systems running real operations for law firms and service businesses. Where there's a measured result, it's on the card; where there isn't yet, it says so.
Projects
Case studies describe a system. These let you drive one. One is a live product you can sign into, one is a shipped product reproduced here on invented data, and one is a reference build. Each card says which it is.
A live B2B SaaS platform for Florida special-education school enrolment. Families chat with Teagan, an AI guide that matches a child to suitable schools on disability type, age, county, scholarship eligibility and services offered; schools subscribe to receive those matched leads. Multi-tenant, bilingual, and billed through Stripe.
Live at teaganai.tech. Three-app monorepo (schools, admin, family), 467 TypeScript files, 61 Postgres migrations, Florida-specific matching rules throughout.
A purpose-built internal CRM that replaced Zoho for a US marketing agency. Client accounts with thirteen service sections, recurring delivery pipelines per service line, two-way SMS synced from GoHighLevel, an AI workspace for search and account health, a visual workflow builder, and every marketing platform the agency runs on wired in behind it.
Built for Found IT Marketing. 790 TypeScript files, 174k lines, 116 Postgres tables across a four-app monorepo.
One console for everything a service business runs on: client relationships and health scoring, client-assigned work, staff capacity and automatic routing, lead generation and campaigns, and every third-party system wired in behind it.
A reference build, not a client engagement — invented from scratch to show the shape of an operations layer.
The lab
Open-source builds where I work out the hard parts — tenant isolation, retrieval quality, reproducible training — before they show up in client work. Every one is public; the code is the claim.
A document-QA backend where every tenant's data is walled off from every other tenant's — the first question any firm asks before handing over their files.
An end-to-end sentiment classifier built the way production ML has to be built — versioned data, tracked experiments, a registered model, and an API at the end of it.
A scripted, templated email dispatcher — the earliest ancestor of the inbox automation I now build for firms.
A recurrent sequence model for climate time series — the problem shape behind any forecast that has to respect what came before it.
A tree-ensemble regressor on messy, real-world pricing features — mostly an exercise in feature engineering, which is where the accuracy actually lives.
How I work
A clear, low-risk path from the first call to a system that runs your operation — and keeps running.
We map the repetitive process you want gone — the triggers, the tools, the edge cases — and agree on what 'done' looks like.
I design the workflow and the AI's guardrails: what it reads, how it decides, and where a human stays in the loop.
I build it on production-grade tooling, wire it into your inbox and CRM, and test it against your real, messy data — not a happy-path demo.
It goes live running your operation. I monitor, tune, and stay on for the months that follow — so it's still working later, not just at launch.
The stack
The toolkit behind systems that hold up on real data — not demo-ware that breaks in week two.
I'm an AI automation engineer who builds systems that run real operations — not demos. I work mostly with law firms and professional-services businesses, automating the repetitive, high-volume work that eats a team's day. I'm a Computer Engineering graduate and a Top Rated freelancer on Upwork, and I care about one thing above all: that what I ship is accurate, secure, and still working months later.
Questions
Still unsure if your process is a fit? The fastest answer is a ten-minute call.
Yes. Systems run in your environment with your accounts and keys — your data isn't sold or used to train public models. I build to the access and retention rules your work requires, with human review where accuracy matters.
A short call to see if a process is worth automating. If it is, I scope a first build with a clear deliverable — usually one workflow you can watch run end-to-end before we expand.
Most engagements are fixed-price per build, or hourly for ongoing work. You'll know the number before we start — no surprise invoices.
That's the whole point of how I build. It's tested on your real, messy data before launch, with a human in the loop where it counts, and monitored after — so the failure cases are handled, not discovered in production.
Law and professional services are my focus, but the same systems fit any business drowning in repetitive email, intake, and document work.
Tell me about the process. I'll tell you straight whether it's worth automating — no pitch, no obligation.
Or email directly · farooqumer694@gmail.com