Software Engineer II · EchoStar · Plano, TX
Sri Sai Srivatsa Yerrapragada
Production AI Infrastructure, Multi-Agent LLM Platforms & Distributed Backend Systems
Software Engineer with 4+ years of experience designing and scaling distributed backend services and AI systems in Python, Go, Java, and C++. End-to-end ownership from RFC to production across multi-agent orchestration (Temporal, Amazon Bedrock), 50M+ vector RAG engines, and multi-GPU PyTorch training pipelines.
Key Benchmarks
Experience
Software Engineer II · EchoStar
Own architecture and RFC process for AI-LC, a durable multi-agent platform built on Temporal and Amazon Bedrock automating the software lifecycle from JIRA ticket to reviewable Git branch with dynamic complexity/budget model routing and persona-based LLM-as-a-Judge quality gates across Dish, Boost Mobile, and Sling.
Software Engineer · EchoStar
Architected an enterprise RAG diagnostics assistant scaling a 50M+ embedding index (Snowflake Cortex to Pinecone) to 1M+ daily queries at sub-150ms P99 latency, cutting manual triage time by 40%, improving Context Recall by 25%+ via recursive structure-aware chunking, and reducing P95 inference latency by 22% via LoRA fine-tuning and async batching.
Machine Learning Engineer (RA) · University of Florida
Engineered multi-GPU PyTorch DDP training pipelines across 8 GPUs for 10M+ hyperspectral samples (-50% iteration time) and a mixed-precision MLflow evaluation framework (-65% evaluation cost, improving spatial-spectral CNN R² from 0.68 to 0.76).
Software Developer · Accenture
Built high-availability Java and C++ microservices processing 2M+ daily transactions at 99.95% uptime and REST APIs serving 100K+ daily requests, reducing latency by 60% via Redis caching and connection pooling.
Technical Specifications
Selected Engineering Projects
Semantic-Search-Analysis
Multi-document semantic search engine over PDF and TXT files with batched transformer embeddings and FAISS L2 indexing.
Cold-Email-Generator
Automated career page scraper and outreach generator matching job requirements against a vector portfolio store.
camera-view-adjustment
Pre-capture active camera view adjustment architecture analysis and multi-view composition benchmarks (FCDB, GAICD, CPC, Pano2Vid).
Movie-Recommendation-System
Content-based movie recommender trained on TMDB 5000 with cosine similarity ranking and live poster retrieval.
Email-Spam-Detector
Real-time SMS and Email spam classifier benchmarking 11 ML models and serving a TF-IDF + Multinomial Naive Bayes pipeline.
NoteBook
Full-stack note-taking web application with JWT authentication, auto-refresh interceptors, and protected routes.