DPIIT
About
I explore the edges of AI and build products that move business metrics.
4 papers accepted at ICML 2026. Co-authored and co-built SIA. My work spans the full arc, from research to engineering to shipping products that create real outcomes.
Current Role
Senior Research Scientist @ HexoLabs
Current Focus
Working toward building an AI Scientist
Research
arXivFeatured
SIA: Self Improving AI with Harness & Weight Updates
A language-model agent that simultaneously modifies task-specific scaffolding and model weights. Achieves 25.1% over prior SOTA on LawBench and 12.4% faster GPU kernels.
Read Paper
ICML 2026
SIA-W: Self-Improving Agents with Test-Time Weight Updates
Autonomous self-refinement via evolving agent structure and test-time reinforcement learning. +16pp on LawBench, -19% GPU kernel runtime.
Read Paper
ICML 2026
Adaptive Proxy Evaluation for Autonomously Improving ML Agents
Addresses the cost/reliability tradeoff in proxy evaluations. MLEvolve achieved SOTA MAE of 0.1354 on MLE-bench within 12 hours.
Read Paper
ICML 2026
Socrates: Structured Questioning Unlocks Latent Knowledge in AI Research Agents
A two-agent system pairing a Scientist with an advisor that can only ask questions. Improved Kaggle test scores on 4/5 MLE-bench tasks with a mean increase of ~56%.
Read Paper
ICML 2026
AIE-Bench: Benchmarking Agents That Build Agents
A benchmark for evaluating whether an AI agent can modify another agent to improve it, covering meta-improvement and self-improvement scenarios.
Read Paper
arXivFeatured
SIA: Self Improving AI with Harness & Weight Updates
A language-model agent that simultaneously modifies task-specific scaffolding and model weights. Achieves 25.1% over prior SOTA on LawBench and 12.4% faster GPU kernels.
Read Paper
ICML 2026
SIA-W: Self-Improving Agents with Test-Time Weight Updates
Autonomous self-refinement via evolving agent structure and test-time reinforcement learning. +16pp on LawBench, -19% GPU kernel runtime.
Read Paper
ICML 2026
Adaptive Proxy Evaluation for Autonomously Improving ML Agents
Addresses the cost/reliability tradeoff in proxy evaluations. MLEvolve achieved SOTA MAE of 0.1354 on MLE-bench within 12 hours.
Read Paper
ICML 2026
Socrates: Structured Questioning Unlocks Latent Knowledge in AI Research Agents
A two-agent system pairing a Scientist with an advisor that can only ask questions. Improved Kaggle test scores on 4/5 MLE-bench tasks with a mean increase of ~56%.
Read Paper
ICML 2026
AIE-Bench: Benchmarking Agents That Build Agents
A benchmark for evaluating whether an AI agent can modify another agent to improve it, covering meta-improvement and self-improvement scenarios.
Read Paper
arXivFeatured
SIA: Self Improving AI with Harness & Weight Updates
A language-model agent that simultaneously modifies task-specific scaffolding and model weights. Achieves 25.1% over prior SOTA on LawBench and 12.4% faster GPU kernels.
Read Paper
ICML 2026
SIA-W: Self-Improving Agents with Test-Time Weight Updates
Autonomous self-refinement via evolving agent structure and test-time reinforcement learning. +16pp on LawBench, -19% GPU kernel runtime.
Read Paper
ICML 2026
Adaptive Proxy Evaluation for Autonomously Improving ML Agents
Addresses the cost/reliability tradeoff in proxy evaluations. MLEvolve achieved SOTA MAE of 0.1354 on MLE-bench within 12 hours.
Read Paper
ICML 2026
Socrates: Structured Questioning Unlocks Latent Knowledge in AI Research Agents
A two-agent system pairing a Scientist with an advisor that can only ask questions. Improved Kaggle test scores on 4/5 MLE-bench tasks with a mean increase of ~56%.
Read Paper
ICML 2026
AIE-Bench: Benchmarking Agents That Build Agents
A benchmark for evaluating whether an AI agent can modify another agent to improve it, covering meta-improvement and self-improvement scenarios.
Read Paper
Experience
Leading AI research and high-impact programs end-to-end, from architecture and novel research to production delivery.
4 papers @ ICML 2026Co-authored & co-built SIALed tech across multiple AI projectsOwned end-to-end delivery for client programs
Built a real-time AI calling system before audio-to-audio models existed. engineered low-latency voice pipelines from scratch and took it from zero to revenue.
$2,000+ MRR within 2 monthsPre audio-to-audio eraUltra-low latency voice AISole engineer on critical infrastructure
Who I've Worked With
Organizations I've worked with across AI research, engineering, and product development.
IP India
Bito
Soliton
Atomicwork
Dashtoon
Get in Touch
[email protected]Yogendra Manawat
Senior Research Scientist @ HexoLabs