Presenting at SmartCom 2025: Generative AI for Automated Video Summarization

I had the privilege of presenting our research paper, "Generative AI-Powered Tool for Automated Video Summarization", at the 9th International Conference on Smart Trends in Computing and Communications (SmartCom 2025), hosted at Crowne Plaza Pune City Centre. The paper is officially published by Springer Nature Singapore in the Lecture Notes in Networks and Systems (LNNS) series.

#The Research & What We Solved

Long-form video - whether webinars, technical tutorials, lecture recordings, or client meetings - is one of the densest mediums to consume. Finding specific information inside an hour-long recording typically requires tedious manual scrubbing or relying on sparse audio transcripts that miss visual context completely (such as code slides, diagrams, and physical demonstrations).

Our work proposed an end-to-end multimodal pipeline combining speech and vision:

• Whisper large-v2 for high-accuracy speech-to-text transcription, achieving 96 percent accuracy on challenging technical vocabulary.
• Video-LLaVA-7B for frame-level visual reasoning, binding speech segments to slide changes and visual cues on screen.
• Structured timestamp extraction that divides recordings into thematic chapters with concise bullet takeaways and key quotes.
• Hyperparameter tuning across beam width, temperature, and learning rates that reduced inference latency by 35 percent while maintaining benchmark accuracy.

#Conference Highlights & Gallery

Presenting at SmartCom 2025 was an incredible opportunity to exchange perspectives with leading researchers, professors, and industry practitioners working on applied AI across India:

#Gratitude to Mentors & Collaborators

This research would not have reached publication without the guidance of my mentors, Dr. Varalatchoumy M and Syed Hayath, whose feedback shaped the methodology and rigorous benchmarking from the beginning.

I also extend my heartfelt thanks to my co-authors: Dhanush C P, Manu R, and Sadhana Vijayakumar for their invaluable contributions and teamwork throughout the experimental evaluation.

#Connecting Research to Production

This academic work on video and speech intelligence directly informed the transcript extraction systems I went on to build in production. Today in Opaeron, the meeting transcript intelligence loop turns messy customer call transcripts into structured action items with owners and deadlines in seconds.

If you are interested in reading how we handle transcript intelligence in production web software: read my engineering note on Meeting Transcripts to Action Items: An LLM Pipeline.