An umpire makesthe call early.

Before the game gets complicated, and based on what is actually in front of them rather than what everyone would prefer to be true. An umpire does not consult the standings before making the call either — which is the part of the name we care about most. Every enquiry receives the same twenty minutes and the same direct answer.

FirstUmpire

~ %cat origin_story.txt

FirstUmpire started from a specific frustration. Agencies tend to sell you a website with no strategy behind it, or “AI” with no working product behind it — and either way the gap gets filled by you, in meetings, translating between three vendors who each own a fragment.

The second was watching well-specified work go unanswered at both ends of the market. A clinic losing bookings to an unattended line, a distributor whose follow-ups never leave, an enterprise team whose integration sits behind a year-long procurement queue — all of them have a defined problem, and all of them are told to come back later.

So we built the platforms first — agents, retrieval, voice, integrations, monitoring — and we integrate them onto your use cases rather than starting from an empty repository. It presents as a service. Underneath, it is the same machinery each time, directed somewhere new. That is what makes a single use case worth taking on and a multi-system programme credible, and it means when something breaks there is exactly one person to call.

Founded in 2022. We are a young company built on years of enterprise work that came before it, operating between Seattle and an engineering team in New Delhi.

~ %

Who runs it

Dev J

devesh@firstumpire.com

Co-founder, product

Product architect — autonomous pipelines, agentic platforms, and protocol safety

Dev builds AI systems by removing the translation layer between product thinking and engineering execution. He architects autonomous multi-agent pipelines and LLM platforms that prioritize safety and structural integrity over surface-level features.

Rather than assembling wrappers, his work focuses on the deep integration of Model Context Protocol (MCP) and multi-model evaluation frameworks. He is interested in how an AI platform actually scales when exposed to real-world usage and unpredictable state.

The first step of his process is diagnosing the actual bottleneck. If an AI deployment is the wrong answer to the operational problem, he will say so on day one rather than shipping code that nobody uses.

PMP® Certified

LLM platform delivery
Agentic systems & MCP
Cross-functional program leadership
AI safety & governance
Product & business outcomes

Naval Tripathi

naval@firstumpire.com

Co-founder, engineering

Engineering leader — data, AI & responsible AI, enterprise platform strategy

As an engineering leader at Microsoft, Naval drives the strategy and implementation of enterprise-scale data and AI platforms that carry analytics and decision-making across the organisation. Thirteen years of leadership across enterprise software, data, and AI sit behind that — and deep expertise in building resilient, secure systems at scale.

He currently wears two hats: running Microsoft's data platform operations, and serving as Responsible AI division lead for MCAPS compliance, where he makes sure internal AI applications meet rigorous standards for safety, fairness, and data governance.

His leadership philosophy rests on three pillars — engineering excellence, culture, and impact. He mentors diverse teams to build scalable, compliant systems, fosters continuous improvement, and aligns technical strategy with organisational vision. He works at the intersection of data governance, AI responsibility, and enterprise architecture, where technology meets trust.

Enterprise data platforms
Responsible AI
MCAPS compliance
Data governance
Platform strategy at scale

What we hold to

Nobody gets ignored

Every enquiry is answered by a founder within a business day — a multi-region rollout and a single overloaded booking line receive the same reply, in the same window. We do not rank enquiries by anticipated invoice, and we do not quote anyone out of the room to protect capacity. Where the requirement is defined, there is a conversation to have.

Honesty over hype

If AI is not the right answer to your problem, we say so on the first call — including when saying so contracts the engagement. That is a refusal of the wrong build, never of the person asking. Frequently the honest answer is a tighter, less expensive first piece, and that is the one we would rather ship.

Ship, then polish

Working software in your hands beats a polished deck in your inbox. A staging environment exists from week one of every build, because anything never deployed is a hypothesis rather than a product. Where the first artefact you need is something to raise or pitch with, that is a deliverable in its own right and we treat it as one.

One team, one thread

You work with one person who holds the whole project, not five vendors each holding a fifth of it. Nothing is subcontracted. The platforms underneath — agents, retrieval, voice, integrations — are ours and already in production, which is what lets us take on a single specified use case and a multi-system integration with the same seriousness.

Talk to one of us

The first call is with a founder, not a salesperson. At any scale.

Book the call

Twopeople,bothstillbuilding.