What kind of work do you take on?

Mostly agentic AI and the systems around it: multi-agent workflows, voice agents, RAG over your own docs, and automations that run in production. I build the product too when you need it, web, mobile, or API. I work in whatever stack you already have.

Do you build multi-agent systems?

Yes, that's the core of what I do. LangGraph, CrewAI, or whatever your team standardizes on. Tools wired through MCP, runs traced in LangSmith, and human checkpoints when something needs approval before it goes out.

Can you build a voice agent that books and hands off to a person?

Yes. On FrontDesk AI the voice agent takes inbound calls, checks a live calendar, books, sends an SMS confirmation, logs to the CRM, and transfers when it gets stuck. Same setup works with Vapi, Retell, or Twilio.

We already use n8n, Make, or Zapier. Can you work in that?

Yes. I've shipped eight n8n workflows with approval steps and proper error handling on FrontDesk AI alone. Make and Zapier are the same idea, different triggers. I fit the tool you pay for, not the other way around.

Do you only do AI, or full product builds too?

Both. A lot of the time the AI sits inside a normal app: Next.js on the front, FastAPI or Node on the back, mobile if you need it. I don't stop at the model demo.

Can you build RAG over our private documents?

Yes. Vector search plus keyword search, reranking, inline citations, and it refuses when your docs don't actually cover the question. That pattern is live today on legal research and clinic document Q&A.

We need more than prompts. What does that look like with you?

Orchestration, tool calls, eval in LangSmith, plus the API and UI around it. Not a chain of prompts in a notebook. I'm a core engineer on a $10M+ funded startup with paying users, so I'm used to shipping, not prototyping.

What clouds and tools have you shipped on?

AWS, Firebase, Supabase, PostgreSQL, Azure when clients require it. On the AI side: LangGraph, CrewAI, Vapi, n8n, and the usual stack. I adapt to what you host on rather than forcing my preferences.

What industries have you worked in?

Legal, healthcare, e-commerce, edtech, SaaS, marketing. The moving parts are similar: agents, retrieval, automations. Domain specifics I pick up quickly from your team.

Are you available, and how do we start?

Open to select projects, remote across US and EU time zones (UTC+5). Send a few lines on what you're building through the form below or email hello@rajajahanzaib.dev. I usually reply within a day.

Raja Jahanzaib / work

case study · 2026

FrontDesk AI

A multi-tenant platform that answers clinic calls, books appointments, and answers staff questions from the clinic's own documents, built end to end.

Solo build, end-to-endAgentic AI · Voice · RAG · Automations

The problem

Clinics lose bookings to missed calls, and staff drown in repetitive questions and manual follow-ups. They needed one system that could take inbound calls, book against a real calendar, answer questions from their own documents without inventing anything, and run the operational busywork around appointments, all across many clinics from a single codebase.

What I built

01
A Vapi voice agent that handles inbound calls end to end: it checks a live calendar, books, looks up patient records, confirms by SMS, logs the transcript to the CRM, and escalates to a human when it isn't sure.
02
A LangGraph multi-agent layer behind a custom MCP server, so tools and sub-agents stay composable and every run is traceable in LangSmith.
03
Hybrid RAG (pgvector + full-text + Cohere rerank) that answers from clinic documents with inline citations and explicitly refuses when the documents don't cover a question, so it never invents guidance.
04
8 approval-gated n8n workflows for the operational load: intake follow-ups, no-show re-booking, review requests, and payment retries, each with error handling and a human-in-the-loop checkpoint.
05
A multi-tenant Supabase, FastAPI, and Next.js foundation, so each clinic is isolated but everything runs on one codebase.

fig.01

FrontDesk AI / system

how it fits together

Voice

Inbound call→Vapi agent→Calendar checkPatient lookupBook + SMS→Escalate to human

Knowledge

Clinic docs→Hybrid RAG: pgvector + full-text + Cohere rerank→Cited answeror refuse

Ops

Events→8 n8n workflows, approval-gatedLangGraph multi-agentcustom MCP server

Platformmulti-tenant · Supabase · FastAPI · Next.js · citations + refusal enforced end-to-end

Outcome

Takes and books calls 24/7 without staff involvement.
Document answers are always cited or explicitly refused, so nothing is hallucinated.
Operational follow-ups run automatically, with approval gates where a human should sign off.

Stack

VapiLangGraphMCPpgvectorCohere rerankn8nFastAPINext.jsSupabaseLangSmith

Want something like this built?

Get in touch

Back to all work