Pari Goyal

My first job. The team handled video across platforms by hand: download it, pull the metadata, write the captions, log it all in a spreadsheet.

I built Claude and n8n agents that ran the whole thing, then spent most of the month on the parts that broke. Rate limits. Tools that disagreed with each other. Logic that had to be split into modules before it held.

GutSense is a handheld breath analyser for gut conditions like SIBO and IBS. Three gas sensors for hydrogen, methane and ammonia sit on an ESP32 and stream over Bluetooth to an Android app in Kotlin.

A Random Forest reads the breath. Clinical threshold rules run next to it, so a prediction on its own is never the answer.

Built with Daisy and Akashdeep Singh.

A premium gifting brand I co-founded. I did the sourcing, the pricing, the vendors and the curation, and worked out which products were worth stocking at all.

A platform connecting startups and corporates. I researched competitors across Crunchbase to sharpen the problem statement, shipped an MVP with no-code tools in a few weeks, and talked to the first users.

Then the founder shut the company down.

Built on n8n, Supabase and the Claude and OpenAI APIs, for live client projects. One pipeline took a lead from a form, checked it against the leads we already had, sorted it hot, warm or cold, drafted a quotation, and sent it to a manager to approve before anything reached the customer.

I also built the SOP system the team ran on. A workflow moves through a set of Markdown files one stage at a time, and each stage states its risks and its pass or fail criteria before anything runs.

A capital advisory firm with no software at all. In three months I shipped two versions of its website, and then the platform the business runs on.

The deal CRM has a public intake form, CSV bulk upload that copes with whatever the columns are called, a Claude summary of every deal, and a hot, warm or cold call on each one. The model drafts. It never sends. Every outgoing email waits for a person to approve it on a single-use link, and the firm's facts live in code where the model can't reach them, so it can't invent them either.

An n8n pipeline reads the news every morning and hands over posts that are ready to go, so daily content became a ten-minute review. I also took the firm from near-zero search visibility to being described correctly in Google and in AI answer engines.

The bug I'd pick if you asked for one: corporate email scanners open every link in an email to check it for phishing. Our approve links were plain GET requests, so the scanners were approving deals by themselves. Approval moved to an authenticated POST.

I was applying to AI engineering roles and getting skill-mismatch rejections without ever being told which skill. The job descriptions were sitting in my browser history the whole time. So I put them in a database and asked them.

ROLEFIT

Retrieval-augmented generation appears in 7 of the 9 rejections and 1 of the 4 that replied.

Before it answers, a grader checks whether the retrieved chunks can answer the question at all. If they can't, it rewrites the query and tries again. After two misses it says the corpus doesn't cover it. That last branch is the point.

ZERO started with one bet: pick the team first, then pick the problem together. I hired founding builders and let them help choose what we build.

I wrote the hiring platform myself in under a week, on Next.js, Supabase, Resend and Calendly. Every candidate gets a single-use link, so a forwarded URL can't take somebody else's interview slot, and row-level security lets an applicant write their own row and read nobody else's.

Eight people across India, in engineering, business and marketing. Everyone writes an update every evening, and anyone can argue with a target in the open.

We're building an assistant on WhatsApp for businesses too big for pen and paper and too small for enterprise software. It answers customers from the business's own records and hands the chat to the owner when it isn't sure. It's still in the making.

Piano. Guitar. Singing. Reading, on purpose now.