Open to opportunities

Vatsalya Soni

@vatsalyasoni

Message

I build reliable agentic AI, RAG, and knowledge-graph systems that automate high-stakes workflows.

India

Message

What I'm looking for

I'm looking to build and research reliable AI systems, especially agentic workflows, RAG, LLM evaluation, and knowledge-graph-integrated applications with measurable real-world impact.

I've built agentic AI systems at Vansun Mediatech, including a six-agent creative-generation pipeline using LangGraph and Temporal that eliminated state-parity failures across orchestrated runs. I also deployed an Azure OpenAI-powered Meta Ads CPL watchdog on AWS EC2, reducing daily manual review time by approximately 97%.

As an AI research intern with the Artificial Intelligence Institute of South Carolina, I designed neurosymbolic architectures for Compact, Custom, and Composite AI Networks through enterprise Knowledge Graph integration. I co-authored three academic papers, including an IEEE-targeted article proposing a Symbolic-in-the-Loop execution gate for autonomous systems.

At Lattice Build Technology, I built a multimodal RAG pipeline for architectural PDFs and CAD assets that reduced retrieval latency by 35%, alongside an OCR-enabled Starling-7B data-entry system that improved extraction precision from 85% to 95%. My work spans LLM evaluation, RAG, fine-tuning, semantic retrieval, and production AI infrastructure.

Experience

Work history, roles, and key accomplishments

Current

AI Research Intern

Current

Vansun Mediatech Private Limited

Jun 2026 - Present (3 months)

Architected a 6-agent creative generation pipeline on LangGraph with Temporal durable execution, eliminating state-parity failures. Engineered an LLM-driven Meta Ads CPL watchdog with Azure OpenAI verdict scoring and one-click Slack execution, cutting daily manual review time by approximately 97%.

LangGraph Temporal Azure OpenAI Systemd MCP Python LLM Slack

Research Intern

Artificial Intelligence Institute of South Carolina

Jul 2025 - May 2026 (10 months)

Designed domain-specific neurosymbolic architectures for Compact, Custom, and Composite AI Networks (C3AN) via enterprise Knowledge Graph integration. Co-authored 3 high-impact academic papers including 2 comprehensive surveys and 1 IEEE-targeted article.

Neuro Symbolic AI Knowledge Graphs Research Academic Writing IEEE

AI Engineer Intern

Lattice Build Technology Ltd.

Jun 2025 - Jul 2025 (1 month)

Architected a multimodal RAG pipeline parsing complex architectural PDFs and CAD assets, driving a 35% reduction in data retrieval latency. Engineered a 45-field automated data-entry system on a fine-tuned Starling-7B LLM with integrated OCR, lifting extraction precision from 85% to 95% and cutting inference costs by 40%.

RAG OCR Starling 7B Fine Tuning LLM Data Extraction Python