Open to internships & new grad roles

Shaik
Kareem

AI/ML & Full-Stack Engineer. I build systems that think, and the interfaces worth thinking in.

I'm a Computer Science (AI & ML) undergrad at MGIT, Hyderabad, working at the seam between machine learning and product engineering. Most of what I build starts as a question — can a browser plan its own workflows? can a power grid tell you it's about to fail? — and ends as something that actually runs, with the pipelines, APIs and UI to prove it.

JavaPythonNext.jsFastAPISpring BootLangChainTensorFlowKafkaMongoDBPostgreSQLDockerReactCrewAIOllamaPrismaFlutterJavaPythonNext.jsFastAPISpring BootLangChainTensorFlowKafkaMongoDBPostgreSQLDockerReactCrewAIOllamaPrismaFlutter
01Selected work

Three projects I'd happily be interrogated about.

The ones from my résumé — each with a real architecture behind it, not a tutorial with the names changed.

01
2025Desktop application

Aura

AI-powered agentic web browser & productivity suite

A desktop browser that can be told what to do instead of clicked through. Aura plans multi-step web tasks, executes them, verifies the result, and carries a full office suite alongside it.

  • Built a high-performance desktop browser with multi-tab navigation, isolated profile partitions (Spaces), collapsible tab groups, split-view layouts and stateful session recovery.
  • Architected an autonomous automation pipeline around a unified ModelRouter for local and remote LLMs, plus a TaskExecutionEngine split into Planner, ActionResolver and Verifier stages.
  • Shipped an in-app office suite with WYSIWYG editors for documents, spreadsheets and PDFs, accelerated by a custom Rust-to-WebAssembly file engine and Tesseract.js OCR.
ElectronReactNode.jsGoRustWebAssemblyGeminiClaudeOpenAIOllama
Closed source — Aura is my own browser and stays confidential. Happy to walk through the architecture.
02
2025Distributed systems + ML

GridSense AI

Predictive grid fault intelligence

Power infrastructure fails loudly and late. GridSense reads thousands of live meter feeds, predicts transformer failures days ahead, and hands field engineers a plain-language work order instead of a dashboard to decode.

  • Engineered a fault-tolerant telemetry ingestion pipeline in Java and Spring Boot with Apache Kafka, validating concurrent time-series feeds from 5,000+ smart meters and 100 transformer units over MQTT.
  • Built a dual-model analytics system — TensorFlow LSTM autoencoders for anomaly detection and scikit-learn Isolation Forest for outliers — forecasting critical transformer failures 48 to 72 hours in advance.
  • Integrated a LangChain diagnostic agent using RAG over a vector database to synthesise maintenance instructions and prioritised work orders for dispatch.
JavaSpring BootApache KafkaMQTTTimescaleDBTensorFlowLangChainFastAPIPython
03
2025Full-stack product

Aizy

Full-stack commerce platform with AI analytics

My largest full-stack build: three separate domains for customers, sellers and admins, on a hybrid database architecture, with recommendation and forecasting models served from a Python side-car.

  • Built an AI-driven e-commerce platform with a Next.js front end, Express.js and FastAPI services, and a hybrid MongoDB + MySQL architecture managed through Prisma ORM and Mongoose.
  • Integrated Random Forest and ARIMA models for product recommendation and sales trend forecasting, exposed to the front end through a dedicated FastAPI layer.
  • Delivered JWT authentication, file upload handling and a responsive interface built with Tailwind CSS and Framer Motion.
Next.jsExpress.jsFastAPIPythonMongoDBMySQLPrismaTailwind CSS

Foreknow

AI cognitive assessment platform

Full-stack platform for early dementia screening — memory, Stroop, image recall and speech tests, scored by a CrewAI pipeline orchestrating Gemini 2.0 Flash, Whisper and Hugging Face models into diagnostic reports and PDF exports.

Next.jsFastAPICrewAIGeminiWhisper

SehatSaathi 2.0

AI health assistant

Azure Health Bot–powered medical assistant with conversational symptom triage, intent routing and protected routes. Rebuilt from the ground up after the first version was shortlisted at SIH; this one reached the CodeUnnati finals.

Next.jsAzure Health BotTailwind CSSVercel

Hand Gestures Controller

Computer vision HCI

Hands-free control of mouse and keyboard using OpenCV for capture and MediaPipe for hand tracking, mapping gestures to real-time input events. Took 3rd place at the MGIT Project Expo.

PythonOpenCVMediaPipePyAutoGUI

Collaborative Text Editor

Real-time systems in Java

Multi-user editor built from scratch on Java Sockets and NIO, with a custom TCP protocol for live text synchronisation and typing indicators behind a Swing GUI. No frameworks, just the network stack.

JavaNIOSocketsSwingMaven

3D Portfolio Generator

GitHub Copilot Challenge entry

Generates animated 3D portfolio scenes from a user's GitHub activity, using Three.js and React Three Fiber with Google OAuth and the GitHub API for personalisation.

Three.jsReact Three FiberOAuthGitHub API

Space Debris Tracker

Orbital mechanics utility

Consumes TLE data and propagates satellite and debris positions over time with Skyfield, with visualisation helpers for analysing conjunctions.

PythonSkyfieldTLEData Processing

DeepSeek for VS Code

Editor extension

A VS Code extension bringing DeepSeek into the editor — command palette integration, activation events and API-driven completions.

VS Code APIJavaScriptNode.js
02About

Two and a half years from hello world to production.

Short version: I like the part where the model meets the message queue.

I started with C in mid-2023, fell for Java a couple of weeks later, and have been shipping ever since. These days my work splits fairly evenly between AI systems (LLM orchestration, RAG, multi-agent pipelines, forecasting models) and the full-stack scaffolding that makes them usable.

I care about the unglamorous parts: message queues that don't drop telemetry, auth that isn't an afterthought, and interfaces that stay fast when the model takes three seconds to answer. Strong DSA in Java keeps the fundamentals honest.

Currently digging into

  • Model serving and inference latency
  • Vector search at a useful scale
  • Production-grade data pipelines
  • Rust, slowly and stubbornly
Education

Mahatma Gandhi Institute of Technology

Now

B.Tech, Computer Science and Engineering (AI & ML)

Sep 2023 – Present · Hyderabad

CGPA 8.72

KMR Junior College

Intermediate, MPC

Jul 2021 – Mar 2023 · Hyderabad

95.2%

AI that ships

LLM routing, RAG over vector stores, multi-agent orchestration with CrewAI and LangChain — wired into real products, not notebooks.

Data under load

Kafka and MQTT ingestion, time-series storage, LSTM autoencoders and Isolation Forest for anomaly detection at thousands of devices.

End-to-end ownership

Next.js front ends, Express/FastAPI/Spring Boot services, hybrid SQL + document stores, Docker and GitHub Actions to get it out the door.

03Toolkit

What I reach for, and why.

Grouped by what it does rather than how impressive the list looks.

Languages

Java is home. Python is the lab.

07
JavaPythonCJavaScript (ES6+)DartHTML5CSS3

Frontend

Interfaces that stay quick under async load.

07
React.jsNext.jsTailwind CSSFramer MotionReact QueryReact RouterFlutter

Backend & APIs

Services, contracts, and the auth in between.

08
Node.jsExpress.jsFastAPISpring BootREST API DesignJWT AuthApache KafkaMQTT

Data & Storage

Relational where it matters, document where it helps.

07
MySQLPostgreSQLMongoDBTimescaleDBVector DatabasesPrisma ORMMongoose

AI & Machine Learning

From EDA to agents in production.

14
LangChainCrewAI (Multi-Agent)RAGPrompt EngineeringLLM IntegrationOllamaHugging FaceTensorFlowScikit-learnWhisperPandasNumPyMatplotlibSeaborn

Analytics & BI

Numbers are only useful once someone acts on them.

06
Predictive AnalyticsExploratory Data AnalysisPower BIAI AnalyticsAnalytical ReportingData Quality Management

Tools & Platforms

The boring stack that keeps things reproducible.

08
GitGitHub ActionsDockerMavenPostmanVS CodeAnacondaAndroid Studio

Foundations

The part that survives every framework change.

03
Data Structures & AlgorithmsObject-Oriented ProgrammingSystem Design Basics
04Recognition

Hackathons, expos, and one very good simulation.

Where the work was put in front of people who weren't obliged to be kind about it.

Finalist — Telangana CodeUnnati Hackathon

Edunet Foundation · Feb 2025 – Mar 2025 · Telangana, India

  • Built SehatSaathi 2.0, a health assistant offering AI-driven symptom analysis, medicine guidance and health recommendations.
  • Paired a conversational interface with a structured medical knowledge base to improve healthcare access in rural and remote areas.
  • Reached the state finals among a large field of entries.

3rd Place — Project Expo, MGIT

Mahatma Gandhi Institute of Technology · 22 Nov 2024

  • Designed and built a gesture-based computer controller using OpenCV, MediaPipe and PyAutoGUI.
  • Enabled real-time mouse and keyboard control through hand gestures as an accessible HCI demo.

Shortlisted — Smart India Hackathon (Internal)

SIH · Sep 2024

  • Created SehatSaathi, a healthcare chatbot offering AI-powered symptom analysis and medicine suggestions.
  • Designed a conversational interface usable from any device, aimed at self-assessment and health awareness.

Tata Group Data Analytics Simulation

Forage · Tata iQ · Oct 2025

  • Completed an AI-powered data analytics and strategy simulation for the Tata iQ Financial Services team.
  • Ran exploratory analysis with GenAI tooling to assess data quality, surface risk indicators and structure insights for predictive modelling.
  • Proposed a no-code predictive modelling framework and an AI-driven collections strategy covering agentic automation, ethical AI and regulatory compliance.
05Journey

The route, roughly in order.

Every language, project and detour that led to the work above.

2023

3 milestones

  1. Jun 2023First steps

    First lines of C

    Started with C, then SQL. The beginning of everything else on this page.

  2. Jun 2023Language

    Picked up Java

    Two weeks in and Java became the language I actually think in.

  3. Nov 2023Fundamentals

    DSA in C

    Started working through data structures and algorithms properly.

2024

6 milestones

  1. Feb 2024Language

    Python, and a treasure hunt

    Learned Python and built a small treasure hunt game with randomised gameplay and voice narration.

  2. May 2024Web

    Moved into web development

    HTML, CSS and JavaScript, then React, React Router, React Query, Next.js and Tailwind in quick succession.

  3. Sep 2024Recognition

    Smart India Hackathon

    Entered SIH with SehatSaathi, an AI health bot. It got shortlisted in the internal round.

  4. Oct 2024Fundamentals

    DSA in Java

    Restarted DSA in Java and kept it as a daily habit.

  5. Nov 2024Recognition

    Hand Gestures, 3rd at Project Expo

    Built a gesture controller with OpenCV and MediaPipe. Third place at the MGIT Project Expo.

  6. Nov 2024Tooling

    Git properly, not just commit -m

    Went deep on Git and GitHub — branching, rebasing, workflows.

2025

8 milestones

  1. Jan 2025Data

    MongoDB

    Learned MongoDB and it became the default store in most projects afterwards.

  2. Jan 2025Recognition

    SehatSaathi 2.0 reached the finals

    Rebuilt SehatSaathi with a new UI and a much larger feature set. Finalist at the Telangana CodeUnnati Hackathon.

  3. Feb 2025AI/ML

    Data science and prompt engineering

    Started on data science, prompt engineering, LLM integration and RAG.

  4. Feb 2025Build

    Started Aizy

    Began my biggest full-stack build: three domains, hybrid MySQL + MongoDB, ML recommendations served over FastAPI.

  5. May 2025Build

    Web scraping, then CollabEditor

    Picked up scraping, then started a collaborative editor in Java and Maven built directly on sockets and NIO.

  6. May 2025Tooling

    Docker and GitHub Actions

    Started containerising things and automating builds instead of deploying by hand.

  7. Jul 2025AI/ML

    Machine learning in earnest

    Moved from using models to training and evaluating them.

  8. Late 2025Build

    GridSense AI and Aura

    Two large builds in parallel: predictive grid fault intelligence on Kafka and TensorFlow, and an agentic desktop browser on Electron, Go and Rust.

Contact

Got something worth building?

I'm looking for internships and new grad roles where AI is part of the product rather than a demo. If you're hiring, collaborating, or just want to argue about whether agents are overrated, my inbox is open.

Hyderabad, India · replies usually within a day