The AI feature is still a vague idea
Turn a broad AI requirement into a focused workflow with measurable value, explicit inputs, and safe fallback behavior.
AI integration development
I integrate Claude and OpenAI where they can shorten a real workflow: drafting, extraction, classification, assistance, or structured generation. Outputs are constrained, validated, observable, and editable instead of being treated as automatically correct.
Problems I solve
Turn a broad AI requirement into a focused workflow with measurable value, explicit inputs, and safe fallback behavior.
Use schemas, application-data validation, retries, and review states so generated output can be trusted appropriately.
Add server-side key handling, usage controls, timeouts, logging, and clear user-facing states for slow or failed requests.
Deliverables
Development process
Define what the model should assist with, what remains deterministic, and how success will be evaluated.
Specify trusted data, schemas, permissions, model boundaries, failure states, and human review requirements.
Implement the complete workflow and test representative inputs, edge cases, latency, cost, and output quality.
Add monitoring, usage limits, fallbacks, feedback loops, and documentation for ongoing model or prompt changes.
Relevant work

Personal Product
A two-sided SaaS for personal trainers and their clients: drag-and-drop workout builders, invite-based two-sided auth, Cloudinary video uploads, engagement analytics, in-app messaging, and AI-generated workout plans. Built solo end-to-end on Next.js 16 and MongoDB to demonstrate real-product patterns, not a demo app.
View case study →
Client Build
Custom WordPress theme for NoteGPT, an AI-powered note-taking app trusted by students at MIT, UCLA, and Purdue. The site covers the full product story — transcription, summaries, flashcards, and multi-format support across audio, video, and PDF.
View case study →
Client Build
A custom WordPress theme built for Lumination, an AI consulting and product studio. Designed to communicate authority at a glance — sharp typography, client logo wall, and a layout built to convert high-stakes consulting leads.
View case study →Client perspective
“Farhan combines technical expertise with a business-focused approach. He doesn't just build features; he thinks about usability, performance, and long-term value. That's exactly why I've brought him onto every company I've founded since: ethical.net, Fritz, Plum Village App, Ecommerce Platforms.”
Catalin Zorzini
Founder · Lumination AI
FAQ
Yes. I choose between Anthropic and OpenAI models based on the workflow, required output, latency, cost, and integration constraints rather than defaulting to one provider.
It can be made appropriately dependable through constrained schemas, validation against real application data, retries, evaluation cases, and human approval for consequential actions.
Yes. I can integrate a focused workflow into an existing product while keeping API keys server-side and preserving the product's current architecture and permissions.
Start a conversation
Send the scope, current state, timeline, and constraints. I usually reply within 1–2 business days.