Vidhya Sudani Vaghasiya, Product lead for two B2B SaaS products, and the engineer who built the AI inside one of them
8 years · Two products, one team · Remote for a Swedish company, 4 years · Surat, India
I ran two B2B SaaS products and built the AI inside one of them. Eight years spanning full-stack engineering and product: rebuilt a failing product in twelve months, defined the model layer that automates it, and shipped four products under my own name.
Two B2B SaaS products in construction technology, owned at the same time, and the AI layer inside one of them. Remote for Sweden, across a distributed team of 20+. Both are behind a customer login, so the links below reach a sign-in page rather than the product.
Rukkor App
Construction collaboration platform
June 2022 to August 2026 · Technical Product Manager
One of the two products I ran, built end to end by the team I guided. Scope, delivery sequence, release strategy and the calls about what got cut all ran through me, from early definition to a product in daily use by construction firms.
Multi-party review, markup over drawings and PDFs, version chains and approval gates, for customers on opposite sides of a contract whose interests do not align. The design problem was whose version counts as the truth when two parties have both marked up the same drawing.
Set the roadmap, sequenced releases, and made the weekly call on how a single team split between Rukkor and Geometra.
June 2022 to August 2026 · Technical Product Manager
Inherited an ageing product carrying enough defect and breakage load that incremental fixing had stopped working. Triaged the backlog, scoped it into versioned releases, sequenced them, and shipped a full rebuild in twelve months with new capability alongside rather than a like-for-like port.
Used by construction professionals for digital measurement and quantity lists that feed directly into what a job gets quoted at.
The rebuild decision was the interesting one: a like-for-like port would have been faster and safer, and I argued against it, because the legacy data model was the thing generating the defects. Porting it forward would have bought a year and then handed the same problem back.
Defined the AI layer inside Geometra before engineering picked it up: the model approach, the fine-tuning strategy, and the dataset creation plan the training rested on.
The dataset was where the product decisions lived. What goes into it defines what the system treats as correct, and in takeoff there is no neutral answer: whether a shared wall counts once or twice, how an ambiguous hatch pattern resolves, what a surveyor would have done. Those are pricing decisions disguised as labelling decisions, which is why they did not belong to engineering.
A wrong output becomes a wrong number on a customer quote, so we gated release on per-class precision against surveyor-verified drawings, not on aggregate accuracy. Averages hide exactly the failures that cost money.
Fine-tuned SAM 2 to auto-annotate architectural drawings and emit XFDF straight into the quotation system. In production with customers.
Taken from idea through build, launch and pricing alongside a full-time role. Nobody scoped any of these for me.
Scanlyz
White-label website audit SaaS
Live · Solo build
Turns any URL into a branded agency report across SEO, performance, accessibility, security and code, with plain language fixes.
Owned end to end: the rules engine, the scan pipeline, pricing and positioning, the launch across Product Hunt and directories, and the domain infrastructure underneath.
The GEO audit runs on Qwen models with prompt engineering, enforced output schemas, and a three-way verdict of verified present, verified absent or unknown, so the system reports uncertainty rather than guessing. Traced a scoring discrepancy between two scans to URL normalisation and per-page logic.
Full two-sided marketplace: coach onboarding, subscription tiers, review and trust verification, AI-assisted matching.
The build is done; the go-to-market is not. I could build the demand side alone but could not seed the supply side without a sales motion I did not have time to run alongside a full-time product role. Knowing that, I would sequence it differently: validate a single vertical with a dozen coaches by hand before building the platform for ten.
The hard problems in this category are supply quality and whether the ranking can be trusted, and both are trust problems rather than engineering ones.
A hyperlocal marketplace for wedding planning in Gujarat, spanning ten vendor categories from venues and catering through decorators, photographers, mehndi artists and baraat. Date-based availability search, wishlists, and an enquiry flow that connects couples to owners directly.
Built and launched; vendor acquisition is the open problem. The directory is live and the flows work end to end, but a marketplace is not a marketplace until supply is on it, and seeding supply is a sales motion I have not run yet.
The decision I would point to is the monetisation model. Vendors pay a low-cost subscription rather than a commission on bookings. With commission you are pushed toward high-value transactions and toward keeping yourself between the couple and the vendor. With a subscription you only win by having the fullest, most accurate directory and by getting enquiries to owners fast, and once a couple has found their vendor you have no reason to stand in the way.
Turns a few answers about business type and required features into a costed, GST-compliant scope and a downloadable PDF. Built for Indian businesses buying their first website, who have no basis for judging whether a quote is reasonable.
The product problem was encoding pricing logic that stays defensible across very different project shapes.
Deliberately not linked. The live landing page carries claims I did not verify before publishing, and I would rather show no link than one I would have to explain.
Client engagements and research from the engineering years, and freelance work since. Listed in full here rather than summarised away.
Araya
Hospital management software for US hospitals
Live · Freelance engagement
Patient self-registration, staff registration for emergency admissions, visit management, patient records, and label and document printing.
Built for the emergency path first: registration has to complete when nobody has time to fight the interface, and a printed label or document has to be right the first time, because there is no undo once it is on a sample or a chart.
Developed a comprehensive business management platform featuring payments, scheduling, marketing, inventory, POS, gift cards, form builder, and instant messaging modules.
Built full-stack features using Python, Django, and React with RESTful API architecture to deliver a seamless user experience.
Available immediately. Four years delivering remotely from India to a Swedish team, with a standing daily overlap with CET. Open to fully remote roles worldwide (direct or via EOR), and to relocation with sponsorship.