AI-Native Full Stack Developer
Building AI-powered systems that automate outreach, attendance, and beyond.
An AI-powered Instagram outreach platform that automates lead discovery and sends personalized messages at scale. Built outreach workflows using Gemini AI, LangChain, Puppeteer, and Docker to reduce manual prospecting effort. Implements context-aware message generation to improve personalization across campaigns.
A live AI-powered mock interview platform that conducts voice-based interviews and provides real-time feedback. Integrated Gemini AI for question generation and response evaluation, and Vapi for voice interface handling. Deployed and publicly accessible.
An automated marketing SaaS platform with wallet-based payments and order processing, live at 9smm.in. Implemented Razorpay payment verification and dynamic pricing workflows. Built end-to-end from product architecture through to launch.
A memory-powered AI assistant that helps users discover local events and offers in their neighbourhood. Built as part of the Neighbourhood Voice product suite using LangChain-based memory workflows. Designed to deliver contextually relevant, personalised local recommendations.
An AI-powered campaign management and email outreach platform with multi-account Gmail integration built for PropertyLedge. Implemented AI-assisted email categorization and draft generation using LangChain and DeepSeek. Also built automated web scrapers and data pipelines to streamline lead generation.
A full workforce management system with role-based dashboards for Admin and Supervisor roles. Includes an AI-powered attendance module using facial recognition, liveness detection, and QR/barcode scanning. Automated payroll processing and reporting workflows by integrating attendance and salary data.
Yeah, pretty actively — Nitesh has been collaborating with teams across India, Australia, and Colombia simultaneously. Coordinating across those time zones while shipping production features is genuinely just part of their workflow at this point.
Nitesh built InstaReach around a LangChain pipeline that handles personalized outreach at scale — it's not just calling an LLM, but chaining together lead discovery, context enrichment, and message generation in a way that actually runs in production via Docker. They've also used LangChain for email categorization and draft generation in the PropertyLedge platform, so it's less "experimented with" and more "relied on to ship real features."
Brezix Studio is a web development and automation agency Nitesh co-founded with friends back in 2024. It started with client work — landing pages, Next.js apps — but Nitesh has been evolving it into more of an AI-driven solutions platform over time.
Node.js and Express.js are the backbone of most of what Nitesh builds, paired with PostgreSQL or MongoDB depending on the project. For managed infrastructure they lean on Supabase a lot, and Docker is standard for anything that needs to run reliably across environments — the facial recognition attendance system at Shree Sai Engineering, for instance, was containerized end-to-end.
Remote is completely fine for Nitesh — they've been doing it already, shipping production work for clients in Australia while based in Pune. As for relocation, that's probably worth asking them directly at niteshdk11@gmail.com.
The platform scrapes a prospect's public Instagram profile — bio, content themes, engagement patterns — and feeds that context into a Gemini AI prompt via LangChain before generating the message. So instead of a generic template, each outreach reflects something specific about that account, which is what actually makes the personalization hold up at scale.
Puppeteer handles the browser automation side — crawling Instagram profiles and extracting lead data — while Gemini AI generates the actual messages and LangChain orchestrates the workflow between discovery and generation. The whole thing runs in Docker so it's consistent and deployable, not just a local script.
Nitesh built in several layers to mimic real user behavior — randomized delays between actions, varied scroll patterns, and rotating session states so the automation didn't produce the kind of uniform timing fingerprint that triggers detection. The goal was making Puppeteer's footprint look indistinguishable from a human browsing session rather than trying to bypass detection after the fact.