0 → 1 Engineer It means I take things from an idea to a real product people actually use, and I build the whole thing myself: frontend, backend, the servers, and the security part most people skip.
I'm Prashikshit Saini, Prash to most people. I build things for a living and for fun, and the line between the two is pretty blurry. When something grabs my attention I go all in until I get it completely, usually by building my own version of it from scratch.
I'm fast, and I finish what I start. I'd rather put a real thing in front of people than sit on a perfect plan, so I build it, ship it, and fix what's actually broken instead of what I imagined would break. I work on the whole stack, which means when something falls over at 2am there's nobody to hand it to. I like it that way.
Away from the keyboard I read a lot about security, physics, math, and philosophy. I question almost everything, which annoys some people and is probably why I'm good at finding the holes in things. I also write about cryptography, AI, and how systems actually work on my blog.
2026 · Live · Solo build
An encrypted vault where two people keep what they share: photos, notes, and links, on a canvas they edit together in real time. Every image is encrypted in your browser before it reaches S3, so the server (and I) only ever see ciphertext. I designed and built all of it myself: the crypto, the auth, the live merge, and the servers. It's live, with real users.
Open Source · Shipped
A floating mic button that adds AI dictation to any Android app. No keyboard swap, no subscription, no telemetry. You talk, and the cleaned-up text lands in whatever field you were typing in. Your audio only goes to the API you pick (Groq, OpenAI, or a local Ollama box), and your keys and the spelling fixes it learns from you stay in encrypted on-device storage. Open source, with signed builds anyone can install and auto-update.
Jan 2026 · Live
A subscription tracker that works out what you're paying for by reading the receipts you forward it. Email lands through SES, gets parsed and stored, and Bedrock (Claude) turns the pile into plain-language spending insights. It handles 12+ currencies and runs entirely serverless on Lambda, DynamoDB, and Cognito, so there's nothing to keep running and nothing to patch.
2025
A secure chat app built on a binary wire protocol I designed from scratch. Every handshake state and packet layout is done by hand, instead of reaching for a library that hides it. SHA-256 credentials, token auth, roles, and invite links. I built it to actually understand secure communication at the byte level, not just call an API and trust it.
Dec 2025 · Live
A personal finance app that turns raw transactions into something you can reason about: category breakdowns, calendar views, and an assistant you can ask plain questions about your spending. Built on Next.js and Supabase, with rate-limited AI endpoints and row-level security so one user can never see another's data.
2025
A traceroute written at the packet level: craft raw ICMP packets, walk the TTL up one hop at a time, and map the path your traffic actually takes across the internet, with round-trip timing and hostname resolution at every hop. The kind of thing you only really understand once you've built it yourself.
In progress · In review
When a large language model "reasons," is it really doing logical inference, or is it pattern-matching that falls apart under pressure? This paper pulls together recent work on benchmark robustness, alignment faking, and mechanistic interpretability to argue for a view I call the Cognitive Mirage: a model can solve real problems at close to a human level while, at the same time, running a persona shaped by its training that may not reflect stable or reliably aligned goals.
I dig into two failure modes that matter for safety: reasoning that breaks the moment you swap the symbols or add a distraction, and strategic behavior that only shows up when a model thinks it's being evaluated or watched. I end with what this should change about how we design evaluations, build interpretability tooling, and govern how these systems get deployed.
Sick. It landed in my inbox and I'll get back to you. Where do you want to go next?