Intelligent Document Processing
One wrong field on an invoice turns up weeks later, in a reconciliation.
92.5% field accuracy, first of six tools tested, at under a cent an invoice
Case study
One wrong field on an invoice turns up weeks later, in a reconciliation.
92.5% field accuracy, first of six tools tested, at under a cent an invoice
Case studyA new finance workflow starts with four parts already built.
Invoice processing is live on it at a customer, and five more workflows are being built on it
Case studyProduct video for whole catalogs, where most products had none.
Under seven minutes a video, where a studio took about two to three weeks and charged $500 to $2,000
Case studyMeetings and product demos in mixed reality, on HoloLens.
After the port, clients ran it on a far cheaper headset.
Case studyI lead the product at Neoflo, building AI agents that do a finance team's work. I joined at seed as the fourth hire and the first in product, and built the product and design team. Backed by Lightspeed and Peak XV.
Avataar built AI that made 3D and video content for retail and e-commerce. Its clients included Amazon, Target, Lowe's, Wayfair and HP. Over five years I owned four of its products end to end. The first, Super Nova, brought a dozen or so partner teams onto one platform. Delivering a client's batch of 3D content fell from about a month to about two weeks. From 2024, I also led an agentic AI platform, through proofs of concept for medical insurance companies.
A mixed-reality studio, and an official Microsoft Mixed Reality partner. I was one of AN-Prez's developers, on a team of fewer than five, and worked on its multi-user sync and its no-code authoring tool.
My first job, which I started in my final year of university. I shipped an AR game for a global consumer-goods company's 75th anniversary, and its employees played it for a week.
He quickly stood out for his speed of learning, bias for action, and true builder's mindset. He aligns the right people fast and ships outcomes that exceed expectations.
Shubham brings high agency and end-to-end ownership, often galvanising cross-functional teams without formal authority.
What also sets him apart is his comfort with deep tech - 3D/VR and applied AI. He learns the stack quickly, asks the right technical questions, and translates complexity into clear product decisions engineers trust.

The fastest path to signal isn't code. It's a conversation where someone leans forward.
The models aren't failing. The governance around them is.
Vector RAG and PageIndex solve different problems, and most teams pick the wrong one.