Venkata Subrahmanyam Upputuri

Architected an enterprise RAG platform (LangChain, GPT-4, Pinecone, MCP tools) to automate extraction and metadata retrieval from 15,000+ regulatory filings, reducing manual compliance review. Built a real-time fraud detection system with PyTorch and SageMaker anomaly models to reduce fraudulent incidents from 1,200 to fewer than 250 per quarter and enable automated financial risk scoring. Deployed a multi-turn conversational AI agent for treasury operations using OpenAI API, LangChain Agents, and FastAPI with tool-use/function calling and sub-2-second latency. Implemented domain-adaptive LLM fine-tuning via LoRA/QLoRA (Hugging Face PEFT) improving F1 from 0.71 to 0.93 across 12 financial NLP categories. Scaled ML serving on Docker/Kubernetes with AWS EKS (auto-scaling, blue-green), reducing inference latency from 800ms to 120ms while sustaining 10,000+ daily predictions. Standardized MLOps/LLMOps using MLflow, Jenkins CI/CD, monitoring, and drift detection to reduce deployment cycles

Developed RESTful and gRPC microservices in Python and Node.js using Apache Kafka for event streaming and Redis caching, supporting 1.2M monthly active requests with 99.9% uptime and sub-50ms average latency. Migrated a legacy monolith to AWS cloud-native microservices (ECS, Lambda, S3) using Terraform, reducing monthly infrastructure cost from $28,000 to $11,000. Built automated ETL pipelines with Airflow and PostgreSQL processing 500GB daily, reducing processing time from 14 hours to 90 minutes. Implemented end-to-end automated tests with PyTest and Jest across 14 microservices, reducing production defect escapes from 23 incidents to 4 per quarter and cutting QA cycle time from 8 days to 2 days. Integrated OAuth2/JWT with RBAC across six services, securing 850,000 monthly sessions with zero security incidents over 10 months. Designed a real-time content recommendation engine using collaborative filtering and PostgreSQL, increasing engagement from 3.1 to 5.8 sessions per week for 420,

Built a tax document process automation tool using Java and Spring Boot, providing REST APIs for document processing/classification; handled 8,000 client tax forms monthly and reduced manual data entry from 40 hours to 3 hours per week, saving $95,000 annually in advisory hours. Consolidated financial data from multiple enterprise client systems via Java and SQL ETL on Oracle/SQL Server with automated reconciliation, accelerating month-end close from 5 days to 18 hours. Delivered AWS cloud migration readiness frameworks for 12 audit clients, identifying $2.1M annual infrastructure savings through rightsizing and reserved-capacity analysis. Automated internal audit workflows via Java backend services, eliminating 1,400 manual reporting hours per quarter and providing real-time dashboard visibility to audit partners across regional offices. Integrated REST API data contracts between third-party financial vendors and internal analytics platform, reducing manual reconciliation from 3 days