Satyabrata

Builder, based in Bengaluru

I build AI systems, then write about why they work.

Two products live in production. An AI pipeline I rebuilt at 12,000 daily users. Thirty-plus articles explaining the parts most people skip.

BugLens, live with billing
IIT Madras, B.Sc. Data Science, 2027

Work

What I've shipped

Production systems with real users and real billing, not weekend demos.

BugLens

Live with billing

AI code review for GitHub · Live, with paying customers

PR reviews are usually slow, inconsistent, or both, and most tools that catch bugs never help a developer avoid the same mistake next time. BugLens runs a three-agent LangGraph pipeline (cross-file impact analysis, security scanning, pattern detection) against the full repository, then grounds its review in a hybrid RAG system that mixes Gemini embeddings with BM25 keyword search over Qdrant. A feature called Shadow Learning tracks a developer's recurring mistakes across pull requests and nudges them privately, with no false-positive noise.

Billing runs on Polar.sh with per-seat pricing, a choice made after watching how engineering managers actually buy tools for a team. Auth is GitHub App OAuth plus Supabase SSR, webhooks are verified with HMAC-SHA256, and access control runs on row-level security.

Node.jsNext.js 16LangGraphGemini 2.5 FlashQdrantSupabasePolar.sh

EnginIQ

Live

AI developer assessment platform

Most technical hiring still runs on algorithm puzzles that have little to do with building an AI product. EnginIQ replaces that with real challenges: fork a starter repo, build a RAG pipeline or a LangGraph agent, submit the link. An automated pipeline checks whether it runs, whether it hits accuracy benchmarks, and whether it follows sound AI engineering patterns, with no human reviewer required. Developers earn a public score and a leaderboard spot; companies post their own challenges and get a shortlist of people who have already proven they can build the thing.

The growth loop is self-reinforcing: more developers show up for the leaderboard, which improves the signal for companies, which funds more challenges, which brings in more developers.

TypeScriptNode.jsNext.jsGitHub APISQL AST Parsing

Experience

Where I've worked

Software Developer Intern

MagicSlides
Nov 2025 to Mar 2026
AI Presentation Platform · Sole intern on the AI pipeline
12K
Daily active users
50%
DAU growth
3
Bottlenecks resolved
  • Traced repeated failures in the image generation service to a single point of failure, then designed a three-tier fallback (primary, secondary, cached) to remove it. Daily active users grew from 8,000 to 12,000 over the following months.
  • Found three bottlenecks that caused cascading failures whenever traffic spiked, and fixed all three without adding infrastructure.
  • Rebuilt the image processing flow to cut latency, which raised both the generation success rate and how often people finished a session.
  • Worked without a senior engineer checking each step: scoped the problem, picked an approach, shipped it, and tracked whether it actually worked.

Technical Content Writer

GeeksforGeeks
Jan 2024 to Dec 2024
Remote
  • Wrote 30+ articles on data structures, web development, and system design, and answered more than 100 reader questions in the comments.
  • Learned to explain a multi-agent system or a webhook security flow to someone who has never touched either, which turned out to be good practice for writing docs and talks later.
Currently open to work

Stack

What I build with

The tools I reach for across frontend, backend, and AI infrastructure.

Languages

TypeScriptJavaScriptPythonSQLGo (familiar)

Frontend

Next.js 16ReactTailwind CSS

Backend & Infra

Node.jsExpressREST APIsWebSocketsSupabase (PostgreSQL)FirebaseVercel

AI / ML

LangGraph (multi-agent)Gemini 2.5 FlashQdrantBM25Hybrid RAGMCP ProtocolSQL AST Parsing

Tools & DevOps

GitGitHub ActionsGitHub App (OAuth + Webhooks)HMAC-SHA256Docker (learning)

How I pick tools

I've made real build-versus-buy calls, not just picked whatever was trending. LangGraph over LangChain for state control. Qdrant over Pinecone for cost and self-hosting. Gemini over GPT-4 for review quality and price. Each one came with a reason I could defend, not a gut feeling.

About

The builder

SM

I like problems nobody owns yet. At MagicSlides I was the only intern on the AI content pipeline, so when the image generation service started failing under load, there was no one else to hand it to. I traced the failure, built a three-tier fallback, and watched daily users grow from 8,000 to 12,000 over the following months. Outside of that job, I've founded two things. BugLens reviews GitHub pull requests with a three-agent pipeline and actually bills customers. EnginIQ scores developers on real AI-building challenges instead of algorithm trivia. Both taught me that shipping is the easy part; deciding what to build, and why, is where the real work happens. I also write: 30-plus articles at GeeksforGeeks, a running habit of documenting build decisions in public, because writing something down is usually how I find out whether it actually makes sense.

2
Live products
12K
DAU pipeline owned
30+
Articles published

Also building in public

Since 2025 I've been posting build logs, AI news breakdowns, and a running series called "Things College Doesn't Teach Us" on LinkedIn and Instagram, mostly documenting the actual decisions behind BugLens and EnginIQ as I make them rather than after the fact.

Ships production systems

Multi-agent LangGraph pipelines, hybrid RAG, GitHub App integrations. Built to hold up under real traffic, not to look good in a demo.

Reasons about tradeoffs

Chose LangGraph over LangChain for state control, Qdrant over Pinecone for cost and self-hosting, Gemini over GPT-4 for review quality and price. Each call is documented, not gut feel.

Owns the outcome

Scoped, built, shipped, and measured a 50% DAU increase at MagicSlides without a senior engineer signing off on each step.

Writes it down

30+ published articles, 100+ reader questions answered, and a running habit of explaining architecture decisions in public as I make them.

Education

Academic background

B.Sc. in Data Science and Applications

Indian Institute of Technology, Madras

Oct 2023 to Aug 2027
Data Science · Machine Learning · System Design · Databases · Full-Stack Development

Languages

EnglishHindiOdia

Building something and need a hand?

Or just want to talk shop. I read everything that lands in my inbox and usually reply within a day.

Get in touch

Location

Bengaluru, Karnataka

Open to work and projects

Available for AI engineering roles and automation projects.

Send a message