Research accepted - IJCAI-ECAI 2026, Demo Track

Kavya Gautam

Computer Science @ Mississippi State University. Building distributed systems, AI pipelines, and research tools. Open to Summer / Fall 2026 internships.

github linkedin
01 // experience

Research & Work

Undergraduate Researcher
Artificial Intelligence Research (AIR) Lab, MSU
Feb 2026 - Present
EV Grid Simulation Platform
  • Research accepted for demonstration at IJCAI-ECAI 2026 (Demo Track); paper: LLM-to-Map: Transparent Conversational Tool Orchestration for Real-Time Multi-Domain Simulation.
  • Re-architected real-time state synchronization from HTTP polling to event-driven WebSockets, eliminating 5-7 requests/sec/client and 2-5s polling delays while pushing network, V2G, scenario, and AI-map updates in real time.
  • Built an agentic simulation interface with 26 OpenAI function-calling tools spanning substation control, V2G dispatch, EV configuration, scenario orchestration, and map control, with live frontend synchronization over Socket.IO.
  • Scaled the simulation platform from ~100 to 700+ concurrent vehicles through WebSocket migration, headless SUMO execution, backend broadcast throttling, and frontend rendering optimizations with 60 FPS vehicle interpolation.
WebSocketsOpenAI APISocket.IOSUMO
Undergraduate Researcher - Software & Data Engineering
Office of Research and Economic Development, MSU
Jun 2025 - Present
NSF SECURE
  • Lead development of a research intelligence platform integrating 8+ federal, academic, and scholarly datasets across 187 Carnegie R1 universities, nonprofits, small businesses, FFRDCs, and federal laboratories.
  • Engineered Python/PostgreSQL ETL pipelines for 400K+ scholarly works and large-scale funding and award data, with deterministic UNITID/ROR entity resolution, source-versioned ingestion, provenance tracking, and validation.
  • Designed a two-tier PostgreSQL architecture separating authoritative source data from derived serving models, enabling reproducible rebuilds, methodology versioning, and cross-entity analytics.
  • Built a TypeScript/React/PostgreSQL analytics application enabling cross-organization funding analysis, peer benchmarking, research-topic exploration, collaboration mapping, and research-security scoring.
PythonTypeScriptReactPostgreSQL
AI Fellow
Mississippi AI Innovation Hub
Dec 2025 - May 2026
Legal Compliance RAG System
  • Scaled a RAG compliance pipeline from a single-state prototype to 7 states and 5,000+ legal documents by building web crawlers for 4 state SOS websites using Playwright and Scrapy, handling interactive and static document sources.
  • Implemented citation-constrained generation and source verification, delivering streamed responses with in-app PDF evidence viewing for auditable legal-compliance answers.
  • Built the retrieval layer on AWS Bedrock and OpenSearch Serverless using hierarchical chunking and S3-backed ingestion across multi-state regulatory corpora.
AWS BedrockOpenSearchPlaywrightScrapy
02 // projects

Selected Work

Distributed Remote Code Execution Engine

↗ github
  • Designed a distributed remote-code execution service with FastAPI gateways, RabbitMQ-backed workers, and isolated Rootless Podman sandboxes; load-tested concurrent execution with Locust at sub-30ms median service latency.
  • Hardened multi-tenant execution with Rootless Podman, network isolation, and custom Seccomp profiles to restrict container privileges and reduce sandbox escape risk.
  • Reduced median response latency by 99% (2.6s → 24ms) via asynchronous I/O, connection pooling, and optimized microservice communication over RabbitMQ.
PythonFastAPIWebSocketsRabbitMQPodman

Vision Transformer for Audio Classification

↗ github
  • Adapted a pre-trained Vision Transformer (ViT) for cross-domain audio spectrogram classification on UrbanSound8K, achieving 71.2% accuracy without audio-specific pretraining.
  • Optimized fine-tuning with Automatic Mixed Precision (AMP), cosine-annealed learning rates, and differential learning rates across frozen and unfrozen layer groups.
  • Trained on an NVIDIA A100 GPU, submitting batch training jobs via Slurm on MSU's shared cluster.
PyTorchViTSlurmNVIDIA A100
03 // skills

Technical Stack

Languages
PythonC++SQLTypeScriptJavaScriptBash
AI & ML
PyTorchscikit-learnNumPyPandasRAG pipelinesvector search (OpenSearch)LLM tool-calling / agents (OpenAI API)
Backend & Distributed
FastAPIWebSocketsRabbitMQPostgreSQLREST APIsMicroservicesasync I/O
Cloud & DevOps
AWS (Bedrock, S3, EC2)Docker / ComposePodman (Rootless)GitGitHub Actions (CI/CD)
Web & Observability
ReactStreamlitPlaywrightScrapyPrometheusGrafanaLocust
04 // leadership

Community

President (formerly Vice President)
ACM SIGHPC - Mississippi State University
Sep 2024 - May 2026
  • Led MSU's ACM SIGHPC student chapter; organized biweekly workshops and coordinated speakers, communications, and logistics to grow the campus computing community, and previously served as VP.