I'm Shreyan Gupta. I design infra systems for AI agents: memory, multi-agent orchestration, and the UX people actually touch. Currently building at 
stealth. Recently shipped 
hermes and 
clawdown. Previously built RAG + compliance AI at 
Exelon.
Based in Philadelphia. Graduating June 2026. Open to full-time roles.
tap to see colorI care about the things most people skip. The edge cases, the impossible deadlines, the parts of a codebase nobody wants to touch. I'm currently shipping my way through building things that matter.
RESEARCH
×
URTC 2025An LLM-Based ETL Architecture for Semantic Normalization of Unstructured Data
EXPERIENCE
STEALTH STARTUPbringing back the attention that AI stole from our brains
DREXEL UNIVERSITY@ Undergraduate Researcher
HANDSHAKE@ MOVE AI Fellow
EXELON@ Data Scientist- ▪automated SPCC risk assessment across 12.5k assets. LLM structuring hit 99.4% accuracy, saved ~$1.2M/yr in regulatory overhead
- ▪built PySpark pipelines processing 30M+ geospatial records with H3 clustering on Azure Synapse for predictive modeling
- ▪designed RAG ingestion for 3.5k docs at ~35ms per file, shipped as Dockerized services on Azure with Terraform + CI/CD
ELAR INDUSTRIES@ Software Developer- ▪built NLP-driven product search with BERT embeddings + XGBoost re-ranking. sub-100ms latency end to end
- ▪shipped prompt-based extraction over OCR'd PDFs at ~87% recall, cutting manual data entry by 60%+
PROJECTS
hermes
ONLINEcontext memory layer for AI coding agents
- ▪built on 5 research papers
- ▪LoCoEval: 96 F1
- ▪drop-in context management for long sessions
mixdora
two-stage fine-tuning pipeline for Mixtral-8x22B (135B)
- ▪SFT on OpenHermes + DPO on UltraFeedback
- ▪DoRA adapters, token-packed training
markdown to styled PDF with MCP server
- ▪instant PDF generation from markdown
- ▪AI agent integration via MCP
LET'S TALK
whether it's a startup, a research idea, or a side project that needs momentum, reach out.