Likith S G
I am a motivated junior at RV University, passionate about computer science and engineering. With a solid foundation in programming, algorithms, and software development, I thrive in fast-paced environments due to my quick learning ability and adaptability to new technologies. Known for strong problem-solving skills and a proactive approach, I excel in collaborative settings and am eager to bring fresh perspectives and innovative solutions to the field of computer science.
Experience
ML Intern @ CVCSI (Center for Visual Computing and Sustainable Intelligence)
Duration: Jun '24 — Aug’24
Developed MPS2U, an advanced dataset condensation method that reduced training time from 4 hours to 5 minutes while maintaining high accuracy.
Optimized synthetic dataset selection using Mutual Information and Reinforcement Learning, improving data quality and model performance.
Engineered and trained ConvNet models for multiple datasets (MNIST, CIFAR-10/100), enhancing preprocessing and visualization pipelines.
Skills
My technical and professional abilities.
Programming Languages
Machine Learning
Web Development
Databases
Tools
Extracurriculars
Languages
Education
My academic background and qualifications.
B.Tech (hons) in Computer Science and Engineering @ RV University
Sep '23 — Present | Bengaluru, India
SGPA: 9.39
Pre-University College @ Reva Independent PU College
Aug '21 — Mar '23 | Bengaluru, India
Stream: PCMC
Percentage: 92.5%
School @ Jain Heritage School
Jun '09 — Mar '21 | Bengaluru, India
Board: CBSE
10th Percentage: 87.4%
Projects
A selection of my work.
Imugi - Agentic AI for Data Analytics
Problem: Non-tech users struggle to analyze raw data without coding skills.
Solution: Built a Gemini-powered web app for natural language data interaction and instant reporting.
Impact: Democratizes analysis, delivering actionable insights in minutes without technical expertise.
Neural Style Transfer for Visual Data
Problem: Traditional style transfer is computationally heavy and lacks semantic understanding of artistic genres.
Solution: Built a lightweight TransformerNet with CLIP-guided loss across 4 transfer complex styles (e.g., Cyberpunk, IMAX) in real-time.
Impact: Delivers high-fidelity, temporally consistent visual stylization on consumer GPUs, suitable for creative apps offering Cinematic-Level color grading features.
Player Re-Identification
Problem: Fast motion and occlusion cause lost identities in sports tracking.
Solution: Created a YOLOv11 pipeline with color/cosine similarity for persistent player re-identification.
Impact: Automates consistent tracking and annotation, reducing manual effort in sports analytics.
Visual Speech Recognition (VSR)
Problem: Audio recognition fails in noise and excludes the hearing-impaired.
Solution: Developed a deep learning model to transcribe speech purely from visual lip movements.
Impact: Enables reliable silent transcription, achieving state-of-the-art 1.0% WER on benchmarks.
Pharmacy Search System
Problem: Standard queries struggle with typos and partial matches, slowing search.
Solution: Built a FastAPI/PostgreSQL backend with Trigram/GIN indexing for instant fuzzy search.
Impact: Delivers typo-tolerant results in <90ms, avoiding slow scans and boosting responsiveness.
Apartment Society Management
Problem: Manual apartment management (tenants, complaints, payments) is inefficient.
Solution: Created a Python/SQLite desktop app with role-based dashboards to centralize operations.
Impact: Streamlines management, automating records and reporting to minimize admin workload.
CodeVault
Problem: Finding structured, practical resources for C, Rust, and Solidity is difficult.
Solution: Built a centralized hub of curated repositories with hands-on examples and interactive contracts.
Impact: Accelerates mastery of systems and blockchain concepts through unified, practical resources.
Bakery Management System
Problem: Manual bakery inventory and order tracking is error-prone and slow.
Solution: Developed a Java backend to automate pricing, inventory updates, and employee records.
Impact: Optimizes daily operations, ensuring accurate stock data and faster customer service.
Process Scheduler Simulator
Problem: Analyzing CPU scheduling algorithms is often abstract, making it hard to grasp key metrics.
Solution: Created a C program simulating FCFS, SJF, SRTF, Priority, and Round Robin algorithms to evaluate performance metrics.
Impact: Provided a practical tool for students to understand and compare scheduling algorithms through hands-on experimentation.
Smart Home Energy Monitoring System
Problem: Monitoring energy usage of individual appliances in real time is often complex and lacks affordable DIY solutions.
Solution: Built a smart energy monitoring system using Arduino and current transformers to track appliance-level energy consumption.
Impact: Achieved 89% accuracy in real-time tracking, providing a low-cost, scalable solution for smart home energy management.
Automated Book Publication Workflow
Problem: The early stages of book publication are often manual, time-consuming, and lack version control.
Solution: Built an end-to-end automated pipeline using the Gemini API to rewrite and refine scraped chapter content, with a Streamlit interface for human-in-the-loop editing.
Impact: Reduced editorial workload by over 70% by automating content generation and review, significantly accelerating the publication cycle.
LMTune
Problem: Sourcing royalty-free music for specific moods is time-consuming.
Solution: Developed a text-to-music generation tool using Meta’s MusicGen to convert natural language prompts into high-quality audio.
Impact: Enabled rapid prototyping of fully produced, genre-specific music tracks in under a minute from simple text prompts.
Collaborative Project Management Tool
Problem: Student teams struggle to coordinate tasks and track progress using fragmented tools.
Solution: Built a full-stack web app with Firebase and React/Redux for real-time task management and team communication.
Impact: Centralized chat, deadlines, and task analytics, reducing coordination time and improving team collaboration.
DPay – Decentralized Wallet DApp
Problem: Managing cryptocurrency often requires complex tools inaccessible to average users.
Solution: Developed a decentralized wallet DApp using Solidity and Vue.js for secure and user-friendly ETH transactions via MetaMask.
Impact: Simplified Ethereum wallet interactions with a GPay-style experience, improving accessibility to Web3.
Voting System – Blockchain-Based DApp
Problem: Traditional voting systems are vulnerable to fraud and lack transparency.
Solution: Developed a decentralized voting system using Solidity and Hardhat to enforce one vote per address and automate winner declaration.
Impact: Delivered a tamper-proof and fully transparent voting mechanism with immutable records on the blockchain.
HeyHolidays – Hotel Template
Problem: Small businesses lack access to professional, easy-to-customize web templates for hotel bookings.
Solution: Created a responsive frontend template using HTML, CSS, and JavaScript with dynamic elements for an engaging user experience.
Impact: Provided a lightweight, open-source starting point for hotel booking websites, simplifying development.
Open Source College Template
Problem: Educational institutions often lack simple, customizable website templates.
Solution: Designed a user-friendly frontend template using HTML, CSS, and JavaScript, modeled after Stanford University.
Impact: Provided an easily customizable starting point for educational websites, reducing development time.
Dataset Condensation (MPS2U)
Problem: Training large machine learning models is time-consuming and computationally expensive.
Solution: Developed a novel dataset condensation script using early stopping and reinforcement learning to create a smaller, yet highly representative, dataset.
Impact: Successfully reduced model training time from 4 hours to just 5 minutes while maintaining 95% accuracy on the MNIST dataset.
Real-Time Sign Language Translation
Problem: The need for accessible, real-time communication tools for the deaf and hard-of-hearing community.
Solution: Built a deep learning model using CNNs and Vision Transformers to classify hand gestures from video frames, trained on a custom-built dataset of over 7,000 images.
Impact: Created a functional prototype for instant sign language translation, demonstrating a viable application of computer vision for assistive technology.
Heart Failure Risk Analysis
Problem: Identifying key predictive factors for heart failure from complex clinical datasets is a major challenge in healthcare.
Solution: Executed a full data analysis pipeline, including data cleaning, preprocessing, and feature engineering, to prepare the dataset for modeling.
Impact: Generated insightful visualizations that clearly identified major and minor risk factors, providing actionable information for clinical assessment.
Certifications
My professional certifications and qualifications.
Interests
A glimpse into what I enjoy outside of code.
- Running
- Building Applications
- Research
- Open Source
- Music
Recently Played
Is There Someone Else? Radio
The Weeknd