Skip to content

I'm Aman Anand

NeuralNoble

Aman Anand

I’m someone who can’t leave a system alone until I understand every moving part in it. AI/ML is the only place where that obsession actually feels useful. I spend most of my time building things that force me to learn small GPTs, vision systems, distributed setups(still exploring), whatever pulls me deeper under the hood. I think in terms of tokens, tensors, pipelines, and agents talking to each other and I like turning messy intuition into something real that actually runs. Long term, I want to build intelligent systems that feel modular, sharp, and alive. For now, I’m just improving every day, experimenting nonstop, and seeing how far I can push this path.

               

Experiences

  • AI Engineer Intern
    @Vedantu • Jun 2025 – Aug 2025

    • Designed and deployed an AI-powered product enabling parents to evaluate their children’s handwritten worksheets by uploading a single photo.

    • Achieved 97% evaluation accuracy (matching or exceeding teacher quality) across test papers.

    • Implemented advanced image preprocessing pipelines including automatic aspect ratio correction, distortion removal, and intelligent cropping.

    • Developed detailed question-wise analytics, annotated answer sheets with teacher-like markings, and knowledge gap identification.

    • Integrated real-time holistic assessment reports for parents, enhancing engagement and trust.

    • Built a pipeline to generate explanatory videos on-the-fly for incorrectly answered questions, significantly improving learning outcomes.

    • Automated content generation tailored to each student’s errors, reducing turnaround time to near-instantaneous.

    • Created a video-based AI doubt-solving system that reduced the percentage of students requesting human teacher intervention.

    • Improved doubt resolution satisfaction scores by delivering more context-rich, visually clear explanations.

  • Machine Learning Engineer Intern
    @VideoVerse • Aug 2024 – Dec 2024

    • Built an end-to-end platform to retrieve highlight clips (sixes, goals, catches, key moments) from thousands of assets via natural-language queries.

    • Created embeddings from OCR, ASR transcripts, and metadata, stored them in a vector database, and implemented hybrid search to improve accuracy.

    • Designed ranking that blends semantic similarity, temporal context and league/team/entity priors; added filters (match, player, team, over/quarter).

    • Ran load testing across multiple services to surface throughput and latency bottlenecks and removed a redundant step in one of the pipeline, reducing end-to-end response time.

Tech Stacks

  • Data Science

                   

  • Dev Tools

             

  • Frameworks

                 

  • Languages