Arjun S. Pai
I build AI-powered developer tools and the full-stack systems behind them — most recently deconstruct.ai, a system that reads an unfamiliar codebase and explains how it works.
Seeking SDE / AI engineering internships & 2027 full-time rolesTechnical Matrix
B.Tech CS at Model Engineering College, expected 2027. I turn ambiguous problems into working software — from AST-based code analysis to email-automation backends. Everything below links to source, because claims should be checkable.
Core
JavaScript, Python, Node.js, Express, React, MongoDB
AI · LLMs
Gemini API (text + vision), structured outputs, prompt architecture, evaluation
Working knowledge
Django, PostgreSQL, Redis, Three.js, Docker, Hugging Face
My Journey
B.Tech Computer Science
Focus on full-stack development, AI/ML integration, and building production software through hands-on projects.
3 flagship projects
Curated by engineering difficulty, not by count. Each shipped solo, end to end; smaller experiments are deliberately kept on a lower tier.
Available for Work
I move fast, pick up new tools quickly, and ship. Currently looking for a team where I can contribute from day one.
What I Actually Engineer
Two flagship systems, shown as pipelines — what the code does at each stage.
The LLM reasons over dependency relationships rather than raw file text, so the generated architecture stays consistent with what the code actually does. Next up: context caching to cut repeat-analysis latency & cost.
Filtering runs before the LLM, so tokens are only spent on messages worth reading — cheaper runs, cleaner digests.
Selected Work
deconstruct.ai
Understanding an unfamiliar codebase takes hours. This makes it minutes.
Not a chat wrapper. deconstruct.ai ingests a full repository, maps how modules depend on each other, and turns that graph into an interactive architecture deck, a performance readout, and a contextual code walkthrough you click through in Monaco. Add real metrics here once measured (e.g. “models a 10k-line repo in Xs”).
View Source →MorningMailAI
Your inbox decides your mornings. This flips the order.
An email digest that runs on its own schedule: fetches the day's messages, filters noise before the LLM runs, and delivers a short, personalized brief — fewer tokens on spam, more signal per morning.
View Source →CodeVault
Your best DSA implementations deserve better than a bookmarks folder.
A private developer portal for storing and returning to DSA code. Auth is built on hashed passwords instead of plaintext or social-only flows, with per-user file isolation so one developer's vault never touches another's.
View Source →More experiments
LeetLog Chrome extension that logs solved LeetCode problems into Notion in one click, replacing a manual logging routine. GitHub → Wellness Wizard Extracts macro-nutrients from a photo of your plate via the Gemini Vision API. GitHub → AirPro Pollution investigation engine that reports environmental data with AI confidence scores. GitHub →Ready to Collaborate?
Placement season is on: seeking SDE / AI engineering internships and 2027 full-time roles. Email is fastest — I read everything.