Aditya Karnam · AI Research Notes
AI Research Explained
AI Research Explained
Making AI papers, experiments, and engineering ideas easier to understand.This section covers my own published research, explainers of other researchers' papers, hands-on experiments, and the engineering ideas that connect them.Coverage
my papers
papers I'm explaining
experiments
engineering notesPapers & preprints3
In progress1+
Citations10
ModePublishing + explaining
Foundational Guides
Start here
Systems-level explainers that teach the fundamentals behind the research below.Guide
What Is an AI Agent Harness?
A Systems Guide to Building and Evaluating Reliable AI AgentsHow models, tools, memory, context, verification, and runtime policies combine to determine agent behavior — with examples from QuECTO, BudgetBench, and AgentABI.Read the guideBuild Series · Part 2
Building QuECTO
From Minimal Agent Harness to Evaluable Coding AgentHow QuECTO grew into a small but measurable coding-agent runtime with reasoning controls, telemetry, subagents, multimodal input, releases, and native evals.Read part twoBooks
AI Systems Design Field Guide
A field guide to model APIs, agent runtimes, MCP, A2A, and production engineering — free to read online.Book · 2026
AI Systems Design Field Guide
A short field guide covering model APIs, agent runtimes, MCP, A2A, and the production engineering that ties them together.Read the field guideMy Papers
Published and in-progress research
Peer-reviewed papers, preprints, and academic work I've authored, with more research in progress.Preprint · 2026
BudgetBench: A Budget-Tiered Protocol and Pilot Harness for Memory Strategy Evaluation in Local Large Language Model Agents
arXiv preprint · arXiv:2609.13149 [cs.LG, cs.CL]An active-budget protocol and reference harness that treats the per-call input-token budget as the independent variable when comparing memory strategies for local LLM agents, sweeping 2K–32K budgets and reporting quality, budget utilization, latency, and budget-violation rates as a first-class outcome.Published · 5 citations · 2025
MEDFIT-LLM: Medical Enhancements through Domain-Focused Fine Tuning of Small Language Models
2025 2nd International Conference on Research Methodologies in Knowledge Management, AI and Telecommunication EngineeringDomain-focused fine-tuning of small language models for medical use cases, evaluated against generic baselines.Published · 5 citations · 2018
Efficient Reduced-BIAS Genetic Algorithm for Generic Community Detection Objectives
Master's Thesis, University of Missouri–St. LouisA genetic algorithm framework for community detection in complex networks that removes structural assumptions and requires no prior knowledge of community count, applied to benchmark datasets and Alzheimer's disease genetic networks.Coming soon
In progress
More papers are in progress; new preprints and peer-reviewed work will be added here as they land.Papers I'm Explaining
Other researchers' work, made accessible
Plain-language breakdowns of papers I find worth understanding deeply, not just my own.Coming soon
First explainer of an external paper is in the works. Check back soon.
Experiments & Engineering
Hands-on work behind the ideas
Practical experiments and engineering notes that ground the research in something buildable.Coming soon
First experiment write-up is in progress. Related engineering work also lives on Field Notes.