Engineering Deep Dives

Sudharsana Rajasekaran

SYSTEMS & ARCHITECTURE 6 entries

Production write-ups on distributed data pipelines, local multi-model LLM agents, self-hosted Kubernetes (K3s + FluxCD), and cloud cost optimization.

Coming soon to Google Play Oct 30, 2026 7 min read

BabbleScribe: Private Voice Typing for Android

I talk faster than I type, and I did not want every sentence going to someone else's server. So I built BabbleScribe: private voice typing for Android that runs on your phone, on your own server, or on a cloud key you bring, with your history encrypted on the device. It is heading to Google Play.

Private voice typingOn your phone, your key or your serverMany languages

0 0

AI Agents & SQL Mar 11, 2026 9 min read

Teaching My Chat App to Query Databases — A Dual-Model Text-to-SQL Agent

How I built a local text-to-SQL agent using two specialized models — Gemma3 for conversation and Qwen3 for tool calling — replacing the tedious SSH-psql-query workflow with natural language questions. Everything runs locally, nothing leaves my server.

Dual-Model RoutingRead-Only SQL Boundary100% Local GPU Inference

0 0

LLM Systems Feb 5, 2026 8 min read

Building an AI Chat System with Practical Memory, Search, and Multimodal Intelligence

A behind-the-scenes engineering deep dive into building a production AI chat system. Memory management, context compression, web search integration, and the infrastructure decisions that make it work.

Rolling Context CompressionLinear Token ScalingHybrid Search + Vision

0 0

Cloud & FinOps Oct 20, 2025 9 min read

Deploying to the Cloud: My Infrastructure Journey

From GCS buckets and expensive load balancers to Firebase's serverless architecture. A journey through cost optimization, architecture decisions, and lessons learned deploying a production portfolio site.

97% Load-Balancer Cost CutServerless Edge HostingAutomated CI/CD Pipeline

1 0

Kubernetes & GitOps Sep 15, 2025 12 min read

Building a Production K3s Blog: Architecture Deep Dive

From a mini PC to a full production blog platform. Deep dive into the K3s architecture, monitoring stack, GitOps deployment, and lessons learned running a self-hosted blog on a homelab cluster.

Self-Hosted K3s + FluxCDPrometheus + Grafana + Loki35W Idle Homelab Cluster

0 0

Systems Trade-offs Aug 4, 2025 5 min read

Two Sites, Two Paths

Firebase for a serverless portfolio versus a self-hosted Kubernetes cluster for a full-featured blog. Why I chose different deployment strategies for different needs, and what I learned from each.

Serverless vs Self-Hosted K3sCost & Control Trade-offsLocal Radeon GPU Inference

0 0