Jaya Vamsidhar Reddy Teramreddygari
Ai Ml Engineer @Truist
Signup · Get unlimited contacts
WORK HISTORY
Ai Ml Engineer @Truist
US
Architected and deployed Truist Assist serving 1.13M+ clients, orchestrating three AI pipelines for real-time data retrieval, policyinquiries, and transactional workflows, reducing agent deflections by 20-25% and topping Curinos containment benchmarks.• Engineered a Neo4j knowledge graph unifying structured banking data and unstructured sources (call transcripts, accounts, prod-ucts) via Cypher pattern matching, cutting AML investigation turnaround time by surfacing hidden money flow relationships.• Built a production RAG pipeline using Docling, LangChain, and Pinecone llama-text-embed-v2 over 200+ policy documentswith semantic search, cutting resolution time from 15 min to 30 sec and reducing inaccurate responses by 25%.• Built a Text-to-SQL system using context-intent embeddings over analyst query history, CodeLlama- 13B via LangGraph andDataHub schema governance, serving 30+ internal analysts with 24% reduction in ad-hoc reporting time.• Developed a ReAct based multi-agent system using LangGraph and AWS Lambda to autonomously handle payments and schedul-ing, reducing operational cost by 35% and improving customer satisfaction by 20% for high-volume banking requests.• Built a real-time voice agent for banking customer support using faster-Whisper STT, Piper TTS, LiveKit WebRTC, LangGraphwith Llama 3.1- 8B, deflecting 40% of inbound calls by autonomously resolving routine inquiries without live agent involvement.• Engineered AWS Lambda functions integrated with RESTful APIs to enable real-time data retrieval within Truist Assist, handlingup to 1 million API calls per month with 99% uptime, ensuring zero downtime during peak banking hours.• Implemented LLM observability and evaluation on Truist Assist using LangSmith, RAGAS, Guardrails AI, and CloudWatch, cutting hallucination rate from 18% to 6% via faithfulness scoring, output validation, and real-time tracing at scale.
EDUCATION
MADANAPALLE INSTITUTE OF TECHNOLOGY & SCIENCE
Bachelor of Technology - BTech, computer science and technology
San Diego State University
Master's degree, BDA
ABOUT JAYA VAMSIDHAR REDDY TERAMREDDYGARI
I build AI systems that work in production, not just in notebooks.With 5+ years designing and deploying LLM-powered pipelines, agentic AI systems, and real-time ML infrastructure in financial services environments where errors carry real consequences. My work sits at the intersection of Generative AI and applied ML: building things that are fast, reliable, auditable, and measurable.What I\'ve shipped:Truist Assist - a multi-agent AI platform serving 1.13M+ clients at Truist Financial Corporation, orchestrating RAG, Text-to-SQL, and ReAct agents via LangGraph and AWS Lambda, handling 1M+ API calls/month at 99% uptime and reducing agent deflections by 20-25%Banking Knowledge Graph - Neo4j-powered entity graph unifying structured banking data and unstructured sources (call transcripts, accounts, transactions) via Cypher pattern matching, cutting AML investigation turnaround time by surfacing hidden money flow relationshipsReal-time Voice Agent - end-to-end voice pipeline using faster-Whisper STT, Piper TTS and LiveKit WebRTC, deflecting 40% of inbound customer support calls by autonomously resolving routine inquiries without live agent involvementProduction RAG Pipeline - LangChain, Pinecone and Docling over 200+ policy documents, cutting resolution time from 15 min to 30 sec and reducing inaccurate responses by 25%Text-to-SQL System - context-intent embeddings over analyst query history with CodeLlama-13B and DataHub schema governance, serving 30+ internal analysts with 24% reduction in ad-hoc reporting timeEnd-to-end MLOps - Docker, Kubernetes, MLflow and Airflow pipelines with FluxCD GitOps, OpenTelemetry, Prometheus and Grafana, cutting deployment cycles from weeks to 48 hours with full drift detection and registry-gated approvalsCore expertise:Generative AI · Agentic AI · Multi-Agent Orchestration · RAG Pipelines · LangChain · LangGraph · Knowledge Graphs · Voice AI · Text-to-SQL · LLM Observability · AWS (Lambda, SageMaker, Bedrock, EC2, ECR) · MLOps · Python · PySpark · SQL · FastAPI · Docker · Kubernetes · Neo4jI\'ve also worked on model risk compliance, bias mitigation for credit scoring (disparate impact ratio reduced from 1.31 to 1.07), and SHAP explainability - because production AI in financial services requires auditability, not just accuracy.If you\'re building AI infrastructure that needs to scale, stay compliant, and actually perform, let\'s talk.
This profile is compiled from publicly available professional sources. Unifers is not affiliated with or endorsed by LinkedIn. Request removal of this profile.