Three at reading size, then the shorter entries. 13 in the full record.
A full-stack SaaS that helps early-career professionals beat ATS filters and land more interviews. Users upload their resume, paste a job description, and get an ATS score, missing keywords flagged, and a tailored resume and cover letter generated in seconds.
Why it existsEarly in a job hunt, you send out a resume and hear nothing back. Most of the time you never find out why. A lot of the time the answer is an ATS filter that dropped it before a person read a word, because it didn't match the keywords in the job description. Nobody tells you that, and the tools that claim to help are either vague or sit behind an expensive subscription. I built CareerKit to close that gap. You upload your resume and paste a job description, and it gives you an ATS score and flags the keywords you're missing. It then generates a tailored resume and cover letter for that specific role. I made it for early-career job seekers, with pricing that works for the Indian market.
careerkit.site ↗
A full-stack Next.js app for searching 13 landmark U.S. Supreme Court cases using two parallel retrieval systems. Quick Search uses Voyage AI legal embeddings and Pinecone for fast semantic2 lookup. Deep Search builds a PageIndex-inspired tree index from case PDFs and uses NVIDIA Llama 3.1 70B to reason over document structure — no vectors at runtime.
Why it existsI wanted to see where vector search breaks. Embeddings find passages that sound like the question; they struggle when the answer depends on where something sits in a document, such as a dissent versus the majority opinion. So I built both and ran the same questions through each: Quick Search is fast, Deep Search is slower and right in ways Quick Search can't be.
semantic-search-rtr1.onrender.com ↗
A browser-based image toolkit to create, edit, and optimize images — including screenshot styling, OG image creation, redaction, background removal, compression, and format conversion. Everything runs locally in your browser with no uploads required.
Why it existsEvery time I needed to compress an image, strip its metadata, or blur out something private, I ended up on a site that wanted me to upload the file first. Screenshots with account details, photos with GPS data buried in them, all sent to a server I knew nothing about. Most of these edits take a few seconds, and they shouldn't need a stranger's server. So I built Snaply. It's a set of image tools that run entirely in your browser, from screenshot styling and OG images to redaction, background removal, compression, and metadata stripping. Nothing gets uploaded, because there's no server to upload to. It's also open source, so you can check that for yourself.
snaply.tanbir.in ↗
A full-stack Next.js (TypeScript) app that lets users create apps and websites by chatting with AI. Features include AI-assisted project creation, user auth via Clerk, background workers with Inngest, a Prisma-backed database, and a modern Tailwind + Radix UI.
codely.tanbir.in ↗
A side project that pings your websites every 5 minutes and emails you the moment something breaks. Add a URL and an email, and it handles uptime checks, SSL expiry tracking, and instant alerts. Built after finding out my own site was down from a friend instead of my inbox.
pingwatch.site ↗
RepoDoc is an AI-powered README generator that streamlines project documentation using OpenAI and GitHub integration. It features live Markdown editing, Supabase-auth authentication, Stripe-powered subscriptions, and a beautiful UI built with Next.js and Tailwind CSS.
repodoc.tanbir.in ↗
A visual tool that demonstrates various pathfinding algorithms3 like Dijkstra's, A*, and BFS in real-time. Users can create walls, set start/end points, and observe how each algorithm navigates the grid to find the shortest path.
pathfinding-visualiser.tanbir.in ↗4 roles, newest first.
Hybrid, Dubai
Remote
Bengaluru, India
Remote
The languages, frameworks and tools I write code5 with, in rough order of how often I reach for them. Most of my work is TypeScript end to end: React and Next.js on the front, Node and a Postgres database behind it, and Python where the AI work calls for it.
React
Next.js
TypeScript
JavaScript
Tailwind CSS
Material UI
HTML
Node.js
Express.js
GraphQL
Sanity
Supabase
PostgreSQL
MongoDB
MySQL
Prisma
Firebase
Python
Hugging Face
Docker
Git
NPM
Vercel
Figma
n8nThe blog7: good things I find online and want to keep or pass on, plus the odd thing I write myself.
A free guide to building a WhatsApp agent with Claude Code and Kapso. About thirty minutes, 2,000 free messages a month, and a few rules you should read before you message anyone.
Hi, I’m Tanbir. I write software for a living, and part of why I work this hard is that I want to buy some farmland one day.
Most of what I write is TypeScript, React and Next.js. Lately a lot of it has been AI work: retrieval, vector search, and getting LLMs to behave inside a real product. That last part takes most of the time. A model can look great in a demo and then fall over the first time an actual user types something into it.
I care about things loading fast, and about them not looking like every other product. I try to keep the code clean, within reason. I have broken plenty of things along the way.
Away from the screen it’s mostly philosophy, physics and history, plus some art and classical music. Football is how I switch off, and I’m a big Cristiano Ronaldo fan. I also watch a lot of films.
Want to work together, or have an idea you’d like to talk through, even a half-formed one? Just ping me. If you have a film I should watch tonight, ping me about that too.
Tanbir Alam
