Gaurav Y.

I build the backend and data layer behind AI products: RAG systems, data pipelines, and self-hosted models running in your own AWS or GCP. Recent work: 10 TB storage benchmarks against AWS S3 Express, a 147M-document MongoDB migration to BigQuery and ClickHouse, and open-source RAG and GPU tooling with video demos in my portfolio below. === WHAT I BUILD & SOLVE === 1. Self-Hosted AI, LLM Serving & RAG Infrastructure • Custom RAG and Model Context Protocol (MCP) servers over AWS S3, PostgreSQL / Aurora, DuckDB, and Lance with strict commit isolation and bounded memory (see "Sab" in Portfolio). • Private LLM and embedding deployments using vLLM (Qwen, Llama 3), Ollama, and Rust / ONNX Runtime microservices (HTTP / Axum & gRPC / Tonic) with dynamic request batching. • Disposable, cost-optimized GPU instances on AWS EC2 (g5/g6 Spot), GCP, and Vast.ai provisioned via Terraform and custom Go daemons (see "Chotu" in Portfolio). 2. High-Volume Data Pipelines, ETL & Object Storage • Large-scale ETL & database migrations: extracted, transformed, and compacted 200GB+ (147M+ MongoDB documents) into Parquet for BigQuery and ClickHouse analytics. • S3-compatible object storage & data lakes: AWS S3, S3 Express, MinIO, Apache Iceberg, Apache Polaris catalogs, and FSSpec integrations. • Storage & GPU I/O optimization: benchmarked object storage at 10 TB TPC-DS scale using Apache Spark and ClickHouse, and built Python-to-Go C-FFI data loaders to prevent GPU starvation during training. 3. Backend Services, Microservices & CLI Tooling • Stateful cluster-provisioning backends, distributed job queues, and REST / gRPC APIs in Go, Python (FastAPI), and PostgreSQL. • Zero-dependency static Go CLI tools and CI/CD readiness/bootstrapping automation for Apache Polaris / Iceberg. === HOW I WORK === • AI-Assisted, Failure-Tested: I use coding agents for implementation speed and spend my own time on system architecture, business logic, and testing failure modes (partial writes, network drops, memory limits, and restart recovery) before shipping. • High Visibility: I turn rough requirements or quick calls into concrete written plans, reproducible scripts, and documented PRs so progress is easy to track. • Schedule & Availability: Monday – Friday with US (PST/EST) and EU timezone overlap. === CORE TECH STACK === • Languages: Python, Go (Golang), Rust, SQL, Bash, TypeScript, Dart / Flutter • Data & Storage: AWS S3, S3 Express, MinIO, PostgreSQL, Aurora, ClickHouse, BigQuery, DuckDB, Lance, Apache Iceberg, Apache Polaris, Parquet, MongoDB, SQLite • AI & ML Infra: vLLM, RAG, MCP (Model Context Protocol), ONNX Runtime, Ollama, PyTorch, Embeddings, XGBoost, LightGBM, YOLO • Cloud & DevOps: Terraform, AWS (EC2 GPU, IAM, S3), GCP, Vast.ai, Azure, Docker, Docker Compose, Linux, systemd, GitHub Actions