Aditya Raj

Experience

21Spheres

Dec 2024 - Present | Remote

GenAI Development Intern

  • Developed an in-house RAG agent from scratch, designed as a pipeline to retrieve and clean data from websites using a web scraper. The system runs llama3.2 locally using Ollama, working as crew_ai agents for information gathering. Working to make a tool which uses web search without any API.
  • Designed and developed a multimodal text-image based chatbot using Ollama and Sentence-Transformers to provide personalized product recommendations and search, with an advanced image similarity search using CLIP implemented using embeddings.
  • Working on a recommendation system evaluation paper that compares different recommendation algorithm techniques and their benchmarking results, proving the win case of one recommendation system over another for a particular type of requirement.

IIT Jodhpur

May 2024 - July 2024 | Remote

Research Intern: SIP@SAIDE 2024

Supervisors: Dr. Bikash Santra & Dr. Dipanjan Roy

  • Implemented image classification algorithms utilizing various architectural models (AlexNet, MobileNet, EffNet, Custom Attention and CNN) on FER 2013 Facial Image Dataset, achieving a maximum accuracy of 53% on the test dataset.
  • Applied signal processing techniques and extracted features using CNN sliding window technique to build a multimodal model by fusing EEG and video data on the SEED V dataset.

IIT Guwahati

June 2024 - July 2024 | Remote

Intern: Dept. of CSE 2024

Supervisor: Dr. Arijit Sur

  • Developed a system in .py to generate a diverse synthetic dataset of geometric shapes and used PyTorch to write a GAN for image completion and Generating Images using the UNet Architecture for Generator.

Freelance

Nov 2024 - Feb 2025 | Remote
  • Developed a clinic management platform hosted on Supabase, featuring RBAC and CRUD functionality, scheduling logic for appointments, optimized SEO, performance, and user experience by implementing lazy loading, debounce, and memoization, achieving LCP under 1 second on slow 4G and under 0.35 seconds on no-throttling (capped at 8MByte/sec).
  • Implemented personalized email notifications with Nodemailer and Google SMTP, created a centralized push-notification system for bulk messaging, and integrated an in-house fully secure OTP service for validation and authentication.