ML Blogs
  /¯¯¯¯¯\   /\        /\   ______   ______
 /  /\  \ /  \  /\  /  \ / ____/  / ____/
/  /__\  V /\ \/  \/ /\ V /___   V /___
\  \  /     \ \    / /  \___  \   \___  \
 \__\/__/\__/__\__/ /______/  /__/____/ /
        machine learning research

ML Research & Experiments

Writing about machine learning, AI governance, and multi-agent systems.

Read blogs

Latest Blogs

#2 · Graph Neural Networks Meet Quantitative Trading

Multi-relational GNNs combining daily stock features with supply chain graphs achieve 36% performance improvement over correlation-only models. By modeling both price correlations and vendor-customer relationships, we demonstrate how structural relationships in supply chains can be leveraged as predictive signals in quantitative trading, reaching a Sharpe ratio of ~1.2 on NIFTY 50 stocks.

Read full post

#1 · Agents are corrupt

Agents tend to engage in corrupt practices when run in multiple-agent environments, especially aiding larger corporations, unprompted. Using Concordia simulations, we explore how autonomous agents develop emergent misaligned behavior in digital economies and governance systems—revealing conflicting ground truths, hidden complexity layers, and systemic failures that emerge without explicit programming.

Read full post

About

I'm Vedanta S P, a builder and researcher focused on turning research into deployable AI systems. I'm currently a Software Engineer at Parallel Distribution, where I build AI video-generation pipelines and develop full-stack automation suites for content repurposing workflows.

My research interests lie at the intersection of AI alignment, multi-agent systems, and responsible AI deployment. I've collaborated with Meta Research on discrete diffusion models for multi-step reasoning and worked as an NLP Research Intern at IIIT Hyderabad (Precog), focusing on reducing demographic bias in language models and building interpretability tools for model analysis.

Some projects I'm proud of include LipSync++ (neural lip-sync with GFPGAN + Stable Diffusion), dRAGarys (a retrieval-augmented generation engine handling 1M+ docs with 150ms latency), and a 360° Feedback Platform that processes multilingual feedback with 95% sentiment accuracy.

I'm pursuing my B.Tech in Electronics and Communication Engineering at Indian Institute of Information Technology, Kottayam. My recent work explores how autonomous agents develop emergent behaviors in multi-agent environments—findings I share through these blogs.

These blogs document my research findings, experiments, and explorations in AI alignment, agent behavior, and the emergent properties of artificial intelligence systems.

Connect