Seven agentic systems, end to end — LangGraph, voice AI, B2B intelligence engines. All deployed.
(Yes, I built all six. No, I don't have a team.)
(each one deployed, not demo-ed)
—Built because no one should need a human to book a meeting.
A multi-agent voice concierge that books meetings, answers FAQs, and routes warm leads — end to end, no human in the loop. Built around a state machine of small, well-defined agents.
—Built because supply chain failures are boring until they cost millions.
A monitoring agent that watches PO, shipment, and supplier health signals across messy systems, then escalates the things that actually matter. Less noise. Fewer 3am calls.
—Built because spray-and-pray cold outreach is everyone's problem.
An agentic outreach system that researches ICPs, personalises first touches, and qualifies replies. Treated the funnel as a graph of LLM calls with clear pass/fail gates between stages.
—Built for IIT Patna iHub. Scored 10/10. Still runs.
A multi-agent recruiting assistant that screens, schedules, and pre-interviews — with a planner-executor router and a full eval harness underneath. Defended end to end; shipped, not shelved.
—Built because losing customers quietly is worse than losing them loudly.
A retention copilot that reads support tickets, billing events, and product usage, then flags accounts about to leave — with the actual reason, not a score. Built for the people who answer the email.
—Built because most forms collect data. This one does something with it.
A form layer that doesn't just collect — it scores every inbound lead against six weighted criteria, writes a reply addressing their actual situation, and decides on its own whether the founder needs interrupting.
—Built because the cancellation email is always the last signal, never the first.
A five-agent pipeline that reads real client emails and calendar events, scores every account's health continuously, diagnoses why a relationship is slipping, and drafts the recovery outreach — before anyone thinks to worry.
I'm Aryan. I started in a B.Sc Math Sciences classroom, taught myself design out of curiosity, and ended up building full AI systems because I was obsessed with the entire problem — not just the spec for it.
Right now I'm co-founding ANR Digital Studios, where I build agentic pipelines from scratch. LangGraph for orchestration. n8nfor the unglamorous glue. Voice AI and retrieval for the surface. No "AI" as a buzzword. Real systems, with real evals, that real people use.
I'm looking for the right AI PM role. Based in Delhi, open to Bengaluru. I care about the problem more than the title — and I'd rather ship the boring version of the right thing than a beautiful demo of the wrong one.
Connect on LinkedIn(the chips are draggable. go on, make a mess.)
Before features, map everything that exists.
Prove the mechanism, not the vision.
Can't improve what you can't measure.
Done beats perfect. Always.
(I mean it. Check the commit history.)
(longer versions live on LinkedIn)
A thread-pull on the difference between sprinkling a chat box on top of a product and actually re-architecting it around an LLM.
Read on LinkedIn →Six systems, one through-line: agents are software, not magic. Treat them like a graph, instrument the edges, and ship the boring version first.
Read on LinkedIn →Spec writing changes. Standups change. The thing that doesn't change is the bar for clarity — if anything, it goes up.
Read on LinkedIn →Open to AI PM roles, product consulting,
and interesting problems worth solving.
Open to AI PM roles and interesting problems.
I check my email obsessively, so you'll hear back fast.
