Venkata Sesh Tej Matta
Artificial Intelligence Engineer @The National Institutes of Health
Signup · Get unlimited contacts
WORK HISTORY
Artificial Intelligence Engineer @The National Institutes of Health
State College, PA, US
Designed and shipped a natural language to TRAPI query-graph compiler(structured JSON) using a 7-node LangGraph agent with Mistral-7B + PyTorch/Transformers; added validation and LLM auto-repair to reach 92% single-hop validity.• Built a hybrid resolver: SciSpaCy + SRI Node-Norm + 7 Million CURIE to Biolink mappings for pinned nodes, and Schema-RAG over Biolink YAML (ChromaDB/Neo4j) for generic nodes and predicates; PyTorch/CUDA embeddings (AMP fp16/bf16, pinned memory, batched I/O) → 40% lower p50 latency and 3× Queries Per Second.• Fine-tuned compact LLMs in PyTorch (LoRA/QLoRA) for predicate/class disambiguation and prompt steering; cut hallucinations around 50% and boosted mapping accuracy vs. RAG-only baseline.• Productionized inference with vLLM (PagedAttention, KV-cache paging) plus 4/8-bit quantization (bitsandbytes, HF Accelerate); deployed as distributed, GPU-first microservices with CI/CD (GitHub Actions), parallel batch processing.
EDUCATION
Indian Institute of Technology, Madras
Bachelor of Technology - BTech, Systems Engineering
Indian Institute of Technology, Madras
Bachelor of Technology - BTech, Chemical Engineering
Penn State University
Master of Science - MS, Computer Science and Engineering (THESIS TRACK)
ABOUT VENKATA SESH TEJ MATTA
Looking for Full Time roles; AI/ML Engineer| GPU| Applied Scientist | Agentic AI | NLP | Knowledge graphs| Health Care|; MS CSE @Pennsylvania State University; B.Tech @IIT Madras; Ex-MLE @Graphene Health AI;
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.