Launch-ready is a claim you have to earn.
I work with founders and with the engineers who own the code, on what the product does today, what it can claim in market, and what has to happen before either is true.
that's me, Shreyan ✦
What I've shipped
all cases →A voice agent that runs at ₹2.29 a minute
Built as backend infrastructure so the client can stand up voice AI customer support for any company that needs it. I costed it layer by layer before writing code, and the stack was chosen against that model rather than against a benchmark.
₹2.29 per minute, all in, at production quality. It is live and running.
₹2.29 at ₹95/USD
The metric was measuring the wrong thing
The reported distress rate was climbing, 21% to 27% year over year, and it read like a market signal. It was 187 spurious labels from companies that had simply not filed their accounts. I traced it to the label definition rather than to the model, and added a reportability gate that holds the rate inside a point.
92.155% on current-year distress flagging ✦ SHAP attached to every prediction
read in detail →₹95,000 a month for bad answers
A facilities management platform was paying a vendor copilot to answer questions about their own data badly. I modelled the unit economics layer by layer before writing code: the cheapest model for intent routing, semantic table selection so only relevant schemas reach the LLM, three tiers of cache.
₹95,000 to ₹35,000 a month ✦ post-deployment spend tracked the model
read in detail →An outside pass before it shipped
An AI coding agent in private alpha brought me in to evaluate the product before wider release. I ran it across six tracks, from static review and security pentest to LLM red-teaming and capability evaluation, each scored against rubrics the industry already recognises, from OWASP LLM Top 10 to CWE, MITRE ATLAS and SWE-bench.
18 findings, ranked and reproducible ✦ the attacks it was built to stop held, the gaps were architectural
read in detail →Partners I bring in
✦ marketing · tech · legalSpecialists I trust for the parts beyond me. I own it for you, so your product ships ready for the market at the standard it deserves.
✦ my partners work end to end under my advisory brief, and I take care of the outcome
how partnering works →Who I work with
Where a company sits on this scale decides what I can do for it. I go deepest with small and mid-sized businesses, close enough to evaluate the product, diagnose what is holding it back, and settle what it can honestly claim in market.
I take these on when the team is highly competent or already structured, where being small is an edge rather than a risk.
My main ground, and where post-seed startups at Series A and B sit too. I evaluate the product, diagnose what is holding it back, and settle what it can honestly take to market.
I help them evaluate the product, diagnose how it is working and where it is heading, and strategise the GTM around their positioning and where they can win customers.
Enterprise runs on field sales, procurement and multi-year cycles. That is a different discipline from the fast-moving work I am built for.
Built for the Opensource
github →cc-habits ↗
open sourcedevelopers across nine coding tools
A tool-agnostic memory layer that learns a developer's habits from real edits and writes them into whichever coding agent they are using, so your context follows you between tools instead of starting cold every session. Hardened against symlink races, Unicode sanitiser bypass and indirect prompt injection.
install from npm ✦ security audit public in the repo
QuantumAI-Noob2Pro ↗
quantum mlA hands-on path through quantum machine learning that is honest about where it helps and where it does not, including problems set so the expected answer is that the quantum method loses to a classical one. Built to learn from, not to sell quantum.
MIT ✦ running since december 2024
Watermark robustness research ↗
adversarialA study of how well text watermarks survive once content is paraphrased and regenerated, so a team can judge whether a watermark is worth relying on. Carries an explicit responsible-use note citing EU AI Act Article 50(2).
Research I've published
google scholar →Articles I've written
all 17 →I run every experiment myself before writing it up.
The dev community
claude developers india ✦ 820+ builders
I founded Claude Developers India, a WhatsApp community of 820+ builders, and I keep it active. I run "Show Your Products, Brainstorm and Validate", online calls where members demo what they are building to each other and get feedback. Each session also brings in a guest speaker or two for a short talk.
I also mentor at hackathons, most recently Code Wizards 2.0 at SRM University, Ghaziabad Campus.
the crew, after Claude event ✎
at NASSCOM ✎
one bug, two heads ✎
mentoring at Code Wizards 2.0 ✎
name tags on, still talking ✎
the hallway track ✎
Want an outside pass over your AI product?
Three questions and I will tell you whether it is worth a call.
Tell me what you need
Pick the closest fit. The questions change with it, and none of them take long. It goes straight to my inbox, nothing else.
an AI product going to alpha or launch, and you want an outside pass before it does