Raj Sharma
Full-Stack & GenAI Developer. Building production AI products that real people use — and the boring parts under them that keep them running.
- role
- Full Stack Associate
- company
- IVY Knowledge Services
- since
- September 2024
- side
- cvforge.in
How I got here
I started where every developer does — somewhere on YouTube, learning JavaScript because a project I wanted to build needed it. That curiosity turned into hands-on work very fast. By September 2024 I was a Full Stack Intern at Ivy Professional School building production AI features for real students. By April 2025, I was a full-time Full Stack Associate leading the AI products I had originally interned on.
Since then I've shipped a live mock interview platform (WebRTC + OpenAI Realtime), an AI résumé builder with ATS scoring, a portfolio builder that turns a CV into a public site with one click, plus the payment, organization-management, and analytics layers underneath them — roughly 111 production REST endpoints, deployed on Microsoft Azure through Bitbucket CI/CD with 99%+ uptime.
On the side I founded CVForge — a privacy-first AI résumé SaaS with an anonymous flow, a JWT-authenticated WYSIWYG editor, comprehensive Upstash Redis rate limiting, OTP-verified referrals, and an admin console that drafts emails with AI and bulk-sends through Resend. Freelance has taken me into custom CRMs (MeraFinanceWala — 3-role RBAC, encrypted backups), bespoke Shopify themes (Wild Safari Quest — 70+ purpose-built sections), and bilingual sites for local businesses (Shivani Events — EN/HI).
The thread through all of it: ship AI features that real people actually use, and make the surrounding product just as solid as the model that powers it.
What I'm into this season
A snapshot of what I'm shipping, learning, and watching — refreshed when it changes, not pretending to be evergreen.
Production AI features
New CVForge editor flows (AI generate-section, recovery codes), the portfolio-builder feature for PrepAI students, and iterative improvements to MeraFinanceWala's CRM for an actively growing client.
Apache Kafka & event streaming
Going deep on distributed event streaming — partitions, consumer groups, fault-tolerant pipelines. Not for a project yet, just because the patterns will be useful when I need them.
Voice agents & MCP
Tracking how voice-first AI products are being built post-Realtime-API, and how the Model Context Protocol changes the way tools are wired into LLMs. Both feel like the next surface area worth specialising in.
Where I'm headed
By 2030, here's the shape I want my work to have.
Run a small, focused AI product studio
Not an agency, not a body shop — a place that ships 3–4 production SaaS products, all profitable, none requiring enterprise sales cycles. Quality over scale.
Be known for voice & realtime AI
The kind of product that feels like talking to a real assistant — not a chatbot with TTS bolted on. I want my name attached to a voice product people actually use daily.
Earn primarily from products, not time
Time should be the bottleneck for ambition, not for survival. Move from time-for-money work toward MRR from products I own end-to-end.
Ship at least one open-source tool people use
Not 50k stars — just a thing that solves a real problem cleanly, that other AI developers reach for, and that I keep alive for years. Quietly important beats temporarily viral.
Mentor 2–3 junior developers
I learned from people patient enough to explain things to me when I was getting started. By 2030 I want to be the person on the other side of that — helping 2–3 juniors ship their first paid SaaS.
Still writing code I want to write
Five years from now I still want to be at the keyboard on the days I want to be — choosing what I work on, not running a calendar that runs me.
What I care about
Ship working, not perfect
The version a real user can paste their résumé into beats the version that has another week of polish. Iterate from production.
Privacy is structural
CVForge never stores résumé content — not as policy, as architecture. Make sensitive data physically impossible to leak.
Cost-conscious AI
Default to gpt-4o-mini. Cache aggressively. Track token usage so cost is visible, not surprising. Most users do not need a frontier model.
Document while building
Notes I take while still in the code are 10× better than the ones I would write three weeks later. README first, refactor later.
Off-screen
Badminton. Frequent enough that it counts as a habit. The kind of break that resets the brain in a way scrolling can't.
Always learning. Whatever tool is the right fit for a project, that's the one I learn next. Most of my best work happened on a stack I picked up while building it.
Have a project in mind?
Open to freelance projects — AI integrations, full-stack web apps, custom CRMs, Shopify themes, and anything that needs both shipped and maintainable.