Agentic AI Engineer & Independent Researcher
Computer Science Undergraduate
B.Tech CSE · Second year · 2026–27
I build agentic AI systems focused on autonomous reasoning, tool orchestration, and reliable execution. My independent research explores AI agent reliability, evaluation, and self-improving systems.
Papers
Examines when evidence of broad AI performance supports an AGI claim. Uses three published cases to distinguish observed results from broader conclusions about intelligence.
Selected work
An evidence-focused platform for exploring unfamiliar source code. Its current implementation ingests repositories and produces bounded structural reports with source citations.
A research workspace for discovering sources, working with documents, and generating cited answers. It brings retrieval, research activity, and supporting evidence into one interface.
A personal agent runtime that coordinates tasks, tools, memory, and document retrieval. It combines bounded execution with human approvals and traceable audit records.
An event-driven driver-monitoring project that combines fatigue signals with traffic context. It connects perception, voice interaction, memory, and intervention workflows.
Research
How can we determine whether an AI agent has genuinely improved from feedback—and whether that improvement makes it more reliable on tasks it has not encountered before?
Independent investigation
Early-stage · Evaluation designI investigate how to evaluate feedback-driven changes in AI agents, with a focus on consistent task completion, error recovery, and performance on unfamiliar tasks.
The planned evaluation compares updated agents with fixed baselines under controlled conditions. It aims to distinguish repeatable improvement from gains caused by narrow tests, changing conditions, or weaknesses in the evaluation process.
Measure task completion, errors, and consistency across repeated runs and different conditions.
Test whether improvements transfer to tasks excluded from the agent's update process.
Use controlled baselines and predefined scoring rules to distinguish genuine gains from misleading results.
Compare an agent that updates its approach using task feedback with a fixed-strategy baseline on held-out tasks. Assess the results using predefined measures and recorded execution outcomes.
Research status: proposed evaluation. No experimental findings are implied.
About
I am a second-year B.Tech student in Computer Science and Engineering. I build AI agent systems and explore how they use tools, manage information, and complete tasks. Alongside engineering, my independent research focuses on agent reliability, self-improvement, and how we can evaluate whether AI systems genuinely improve over time.
I explore how AI agents are designed to plan tasks, use tools, manage memory, and coordinate actions. My focus is on building structured agent systems that are useful, maintainable, and reliable.
I am interested in how to test whether AI agents perform reliably, recover from errors, and improve through feedback. My research focuses on measuring these changes and checking whether improvements hold across different tasks and conditions.
I am interested in how AI systems could improve their own capabilities through repeated cycles of evaluation, feedback, and updates. I want to understand how such improvement can be measured, what limits it, and how we can distinguish genuine progress from misleading results.
Contact
I welcome feedback on my projects and research direction, and conversations about AI systems and engineering collaboration.