Case studies and research on multi-agent systems, credit infrastructure, and AI that has to explain itself — written from products I have helped build.
Four questions the work keeps returning to.
Orchestration, memory, and human-on-loop review — what it takes for agents to work inside systems people already depend on.
One AI at HROneAuditability, evaluation, and review paths. How a system earns the right to act on its own, and how you know when it has.
The One AI JourneyDecisioning for people the bureaus cannot see. Default is not one thing, and underwriting it as one thing is where systems break.
Structuring RiskProduct craft under India-scale constraints, where the chaos of the market is the specification.
India: The Ultimate Product LabThree products, each written up as the call that was made and the evidence that followed. Full case notes coming soon on Substack.
How do you build AI that understands context across payroll, attendance, leave, and performance — without hallucinating on compliance-sensitive data?
Running in production across live enterprise orgs.
Full case note on Substack — coming soonHow do you build a credit engine that works without CIBIL, for borrowers the bureaus cannot see?
Scaled from pilot to a pan-India lending book.
Full case note on Substack — coming soonMahindra TEQO managed a gigawatt-scale renewable portfolio. Can we predict inverter failures before they happen?
Deployed across the managed renewable portfolio.
Full case note on Substack — coming soonThe real skill of a PM is simple: can you take something blurry and make it clear enough that execution feels effortless?
They are not just selling frames. They are making a broken process frictionless, and profitable. What the Lenskart IPO tells us.
Search rewarded articulation. AI rewards vagueness. We are building browsers that do not let you get lost.
New essays by email: subscribe on Substack
I am Associate Director, Product & Design at HROne, working on AI products and platforms. Before that, credit infrastructure at Paytail, work-pattern analytics at Soroco, and renewable energy operations at Mahindra. The cases above come out of that work.
Full work historyI teach GenAI product development at Masters' Union and ISB Executive Education, and speak on AI systems and product craft.
Teaching and talksOpen source: public contributions to LiveKit Agents, Kill Bill, Zerodha Kite Connect, and Netflix concurrency-limits.
github.com/sir-adtetris.codes: local context compression for coding agents. 9K downloads.
tetris.codesNexus Prime: local-first memory and orchestration for AI coding agents. 23K+ downloads.
npmFounders and operators I worked with, on what came out of it.
"Adarsh doesn't just advise — he architects. When we were struggling with our AI roadmap, he came in and restructured our entire approach. We went from a confused chatbot to a multi-agent system that actually worked in production."
"Most advisors give you frameworks. Adarsh gave us an operating system. His work on our recommendation engine wasn't just 'consulting' — it was building the foundation that helped us scale."
"What sets Adarsh apart is his ability to think at the intersection of systems and scale. He helped us design our community growth loops when we were just starting. The playbook he built is still running today."
The full case notes, and the essays that work through the same questions from the outside.
Reachable at hello@adarshagrahari.com or via connect.