Product Manager at Actabl
I build B2B SaaS products, and the AI features that take manual work out of customers' day.
8 years in product and 13 in tech, with 10 of those 13 years delivering for U.S. clients and employers. At Actabl, a U.S. B2B SaaS company, I own the web and mobile products and set the roadmap with the C-suite and key customers. I've shipped AI since 2020: machine learning, then vision and document AI, and now an LLM product of my own.
Selected work
-
Redline: an LLM that has to show its source
A web app that reads a client agreement for a freelancer and flags what they're about to sign. Every flag shows the exact sentence it came from. Code checks that sentence word for word against the document, and a flag whose sentence doesn't match is never shown.
-
AI Asset Setup: from a photo to an asset record
Customers photograph a piece of equipment and AI pulls out details such as the equipment type and serial number to create the record. I owned it end to end, and we piloted it to measure real-world cost and accuracy before opening it to every customer.
"We reduced asset onboarding from 30 days to under a week."an Actabl customer, quoted on actabl.com/hotel-ai
-
Associate Engagement: turning routine maintenance into measurable work
A module that turns routine hotel maintenance into measurable achievements, which I took from idea to launch.
Sites using it ran 2–3x higher scheduled-item compliance than sites without it.Actabl blog, April 2025. A comparison between sites, from data I pulled.
Shipping AI since 2020
Each step moved closer to the model itself: classical ML in a bank-facing platform, then vision and document AI inside a SaaS product, now an LLM product where the grounding, the evals and the judge are mine to design.
- 2020–21Machine learningAt Traydstream I owned the AI/ML roadmap for a platform that global banks used to check trade documents. I improved model recall through real-world data validation loops with Data Science, Engineering and key customers.
- 2025–26Vision and document AIAt Actabl I owned AI Asset Setup end to end, piloting it on real customer photos to measure cost per asset and accuracy before general release.
- 2026LLM productRedline: grounding enforced in code, an LLM judge whose disagreements are logged rather than gating, and a health metric built to catch a model that goes quiet.
Day to day I work AI-native. I use Claude Code daily to create PRDs and specs, prototypes, data analyses, diagrams and launch docs, and I've written my own skills, hooks and subagents for it. I also work in Codex.
Where I've worked
Actabl · Product Manager · 2023–now
U.S. B2B SaaS. I own the web and mobile products hospitality businesses use for asset maintenance and operations, and set the roadmap with the C-suite and key customers, on a platform that serves 10,000+ properties.
Xiaomi India · Product Manager, Consumer Payments · 2021–23
Mi Pay, a UPI payments app with 10M+ registered users. I grew daily transaction volume 40% by launching and iterating on core features: bill pay, onboarding experiments, and the merchant and partner integrations.
Traydstream · Senior Manager, Product · 2020–21
UK B2B SaaS for trade finance. Owned the AI/ML roadmap and the client accounts behind most of the company's revenue.
Deloitte USI · Product Consultant · 2015–18
Digital transformation for U.S. P&C insurance clients: a digital policy issuance platform and risk tools for premium leakage and claims losses.
LTI · Test Engineer · 2012–15
QA on a large U.S. insurer's policy systems. Where I learned how software fails.
SPJIMR · PGPM, Information Management · 2018–19
Ranked 74th globally, Financial Times 2026. I left a stable job for it to move into product for good.
How I work
Watch the work before designing the fix
AI Asset Setup came from watching customers photograph equipment and then retype the details at a desk. The fix was already in their hands.
Put the guardrail in code, not in the prompt
In Redline every flag carries the exact sentence it came from, and code checks that sentence word for word against the reader's document before the flag is shown. A prompt asks the model to behave. A check doesn't depend on it behaving.
Fix the mechanism, not the incident
When something breaks I add the check that was missing: a pilot gate before general release, or failsafes so one failed integration can't block a transaction.