Honestly reported.

Instill, the Culture Operating System

An early-stage company set out to build AI that helps enterprises understand their own culture: how teams meet, communicate and perform. It had research, ambition and engineers. It had never had a product person. For two and a half years, I owned product, programme and governance.

• Shipped the beta in three months with eight people across five countries, then turned 100+ discovery interviews into ten-plus shipped features.
• Fine-tuned a domain LLM on fifty years of behavioural research, with curation standards and quality gates that raised accuracy 40%.
• Wrote the Instill Flow Bias Audit: fairness testing across transcription, speaker identification and sentiment, adversarial testing for prompt injection, thresholds that force investigation and remediation, and an ethics escalation path.
• Drove the commercial engine: multi-tenant architecture across Azure and AWS, onboarding that lifted weekly actives 80% and usage 300%, and GTM that cut acquisition cost 25%.

Lloyd’s Register, an AI strategy that priced itself

One of the world’s oldest maritime services organisations wanted to know where AI genuinely belonged in its commercial operations, and what it was worth. Institutions like that do not adopt AI on vibes; the case has to survive scrutiny.

• Built a three-phase deployment roadmap grounded in the Playing to Win framework, sized against the sales operation.
• Put value math on every recommendation: forecasting accuracy lifted from the 60 to 70% range toward 85%+, three to five hours a week returned to each rep, and 2 to 3% margin from pricing optimisation.
• Kept humans in the loop wherever decisions carry weight, and stayed honest about which numbers are targets rather than promises.

Vizara, heritage in three dimensions

Vizara digitises India’s cultural heritage: monuments laser-scanned, reconstructed, and brought back to people through 3D printing and extended reality. Culturally priceless, commercially unproven. Over four years I worked every layer, from the print bed to the P&L.

• Built the showpiece: the largest 3D-printed collection of UNESCO World Heritage Site replicas, with five engineers, in materials 1.5x stronger at half the cost. Shown at the National Museum and Dubai Expo 2020; publicly recognised by India’s Prime Minister and Commerce Minister.
• Grew the business line 40% in six months, built a $150K-a-year e-commerce channel, then as product and BD manager launched eight XR products in eighteen months with revenue up 90%.
• Made delivery boring: agile milestones cut time-to-market from four months to three and raised on-time delivery from 60% to 95%.
• Signed the institutions: a subscription model, five multi-year contracts including the National Museum of India, and a $400K+ pipeline across India, the UAE and Singapore.

needl.ai, growth with the guesswork removed

needl.ai turns scattered enterprise information into a single searchable layer. The product worked; the funnel leaked. My year there was a masterclass in finding out exactly where, and fixing it in shippable increments.

• Rebuilt the signup funnel through fifteen A/B tests across 50,000+ users: drop-offs down 20%, 5,000+ new users a month, feeding 30% MRR growth.
• Read 10,000+ user sessions to find where onboarding lost people; eight UX fixes took completion from 40% to 85% and lifted D30 retention 15%.
• Shipped an AI search filter that hit 15% adoption in its first quarter and moved NPS up eight points, on the back of four analytics-backed PRDs.

AfterWard, a recovery companion for surgery patients

After surgery, patients go home and largely disappear from view until something goes wrong. With four teammates, Ulya Khairani, Ronit Jitesh, Irene and Harley, I built the concept for an AI agent, voice and text, designed around three surgical pathways in NHS Edinburgh: proactive calls, recovery checks, triage grounded in clinical guidance, and escalation to a clinician the moment a patient needs human eyes.

• Judges at Edinburgh Innovations’ Startup Fast Track named it Most Viable Business.
• It fused my two instincts: in healthcare, the escalation path is not an edge case. It is the design.

Two education ventures, run on conviction

AI tutoring products start in English; hundreds of millions of Indian students do not. Both ventures were built at night, beside a demanding day job, and neither was meant to make money. They kept my hands on the tools and my standards honest.

• ShikshaSathi, supported by Ola: a Hindi-first AI learning platform for the national curriculum. Llama 3 fine-tuned on textbook content, with an answer-validation framework so the AI was checked before a student relied on it.
• AceItChamp: an AI revision and assessment platform with adaptive study plans, piloted in three coaching institutes and iterated weekly on student, teacher and analytics feedback.

Oundle Town Council, a roadmap for a market town

A historic English market town wanted more visitors and a stronger local economy. I directed the engagement: competitive benchmarking against comparable towns, primary research with twenty-plus stakeholders, and a five-year roadmap covering tourism positioning, visitor experience and economic development. Local government moves on consensus, and earning it is a skill no product job teaches.

Effectus Research, governance turned into a working product

My current internship alongside the MBA: working with founder Brian Mooney to translate frameworks from corporate governance and strategic risk into a practical decision-quality tool for senior leaders. Boards commit to strategies constantly and rarely write down why, so when a market assumption quietly stops being true, nobody notices until it is expensive. I built the software for that: the Strategic Validity System.

• Every commitment is scored on five weighted parts, the CARMR framework: what you are committing to, the assumptions it depends on, the reasoning that links them, whether your terms mean the same thing to everyone, and how it gets reviewed, rolling into a single 0 to 100 Commitment Integrity Score.
• Disagreement stays in the record. A minority position that contradicts the stated assumptions is preserved permanently, never resolved by vote or quietly dropped.
• Language drift is caught automatically, measuring how far a term's stated definition has moved from how it is actually used elsewhere in the record, and flagging it before that ambiguity becomes an unnamed disagreement.
• Nothing is overwritten. Every edit, activation, or review is saved as an immutable, versioned snapshot of what the team believed at that point in time.