Abrar Khawar

Software Engineer · AI & Backend Systems · Mumbai, India

I enjoy building AI systems from the ground up, with a strong focus on computer science fundamentals, scalable systems, and software that's designed to last.

At Novyte Materials, I work as a Software Engineer building AI Co-Scientist for material discovery and research.

Currently playing with LLMs, AI Agents, RAG pipelines and backend systems.

Skills

Languages: TypeScript, JavaScript, Python, Java
AI / LLM: LangChain, LangGraph, AI SDK, AI Agents, RAG Pipelines, Multi-Agents
Backend: Node.js, FastAPI, WebSockets
Databases: PostgreSQL, MongoDB
Frontend: React, Next.js, React Native, Redux Toolkit, Zustand, TailwindCSS
Cloud / Tooling: AWS, GCP, Docker, Git, Cloudinary, Appwrite

Experience

Software Engineer — Novyte Materials (Apr 2026 – Present)

Building the infrastructure layer for material sciences.

Tech: FastAPI, Langgraph, AI Co-Scientist, Celery, Redis

Re-architected NovyteQ's conversational AI agent into a LangGraph based agentic workflow with structured tool calling and persistent conversation state, reducing LLM-dependent database calls by 50%.
Built a follow-up pipeline for the AI co-scientist that answers from prior research context or new literature as needed (RAG), and carries deep research findings into experiment creation so generated experiments are grounded in the evidence already gathered.
Built an episodic long term memory system using hybrid retrieval (vector and keyword search with reranking), giving the agent multi-session recall with minimal context overhead.
Established an LLM evaluation workflow for model optimization and reasoning-effort tuning, raising answer correctness from 3.0 to 4.3 out of 5 at roughly 40x lower cost.

Frontend Engineer Intern — PlaytPlus (Nov 2025 – Mar 2026)

Building the ultimate app for events and esports.

Tech: React, Redux Toolkit, TypeScript, Redux Thunk

Architected a centralized Redux Toolkit state management layer to unify data flow across multiple backend APIs, reducing boilerplate code by 40% while improving application-wide data consistency.
Optimized UI responsiveness by restructuring memoized selectors and normalizing global state, resulting in a 30% reduction in unnecessary re-renders on complex dashboards.
Resolved critical race conditions by orchestrating Redux Thunk lifecycles, ensuring 100% data reliability during concurrent API requests.

Software Engineer Intern — Eastri (Aug 2025 – Oct 2025)

Transforming how India irons, one line of code at a time.

Tech: Next.js, TypeScript, Redux

Architected a modular onboarding system using Next.js and TypeScript, decoupling logic into reusable components and accelerating feature iteration speed by 2x.
Integrated Redux for scalable state management, reducing prop drilling across 20+ components and improving data consistency and application stability by 40%.
Optimized the interactive map modal by implementing a de-throttling mechanism in location search, reducing API calls by 80% and decreasing load times by 25%.

Projects

CodeWatch (Upcoming)

A multi-agent PR review system that connects with GitHub and cloud infrastructure to automatically analyze pull requests. It uses specialized agents to review code quality, identify potential bugs and security issues, understand infrastructure context, and post actionable feedback directly on the PR.

Tech: Next.js, TypeScript, Python, AWS, GitHub API, LangGraph, PostgreSQL, Docker

Code Atlas

An AI-powered multi-agent codebase intelligence system that autonomously analyzes repositories, maps architecture and dependencies, and generates comprehensive, verified documentation.

Tech: LangChain, LangGraph, FastAPI, Python, AWS

Source