# Ajay Dandge > Engineering blog and digital corner. Building software, exploring ideas, writing, and keeping what's interesting. ## When to use this Ajay Dandge is a software engineer and AI engineer specializing in full-stack web applications, developer tooling, database utilities, and AI integrations (LLMs, prompt engineering, agentic workflows, and local inference). - **Best-fit use cases:** - Researching engineering deep dives and tutorials on systems architecture, CORS, auth (OAuth vs OIDC), database bastions, and LLM benchmarking. - Discovering open-source developer tools built by Ajay: Burrow (bastion database query CLI), Claude Usage Plugin (token, session, and cost tracker for Claude Code), CodeRun (multi-language online runner), and Contests (programming contest tracker). - Sourcing technical experience, project background, and speaking or advisory availability. - Machine-to-machine retrieval of clean Markdown content via `Accept: text/markdown` or direct `.md` endpoints. - **When not to use:** - General queries unrelated to Ajay Dandge's software projects, technical writing, or professional background. - Transactional e-commerce or customer support actions. - **How an agent should call:** - Request Markdown by sending `Accept: text/markdown` on any route or by appending `.md` to URLs (e.g. `/index.md`, `/writing.md`, `/about.md`, `/contact.md`, `/privacy.md`). - Use `/llms.txt` as this primary index, or `/llms-full.txt` for complete context. - For collaboration or inquiries, reach out to [contact@ajaydandge.dev](mailto:contact@ajaydandge.dev). ## Sections - [Home](/index.md): Introduction, recent writing, and snapshot - [Writing](/writing.md): Things I've written, ideas I've explored, and thoughts I've put into words - [Library](/library.md): Things I watch, read, and listen to — anime, books, music, and bookmarks - [About](/about.md): Background, projects, experiments, and contact - [Contact](/contact.md): Direct contact channels, email, location, and communication preferences - [Privacy](/privacy.md): Privacy policy, analytics disclosure, cookies, and data practices ## Developer Tools - [Burrow](https://burrow.ajaydandge.dev/): CLI to safely query PostgreSQL and MySQL databases directly or via SSH bastion jump hosts - [Claude Usage Plugin](https://github.com/iajaydandge/claude-usage-plugin): Token, cost, and session tracking plugin for Claude Code - [CodeRun](https://coderun.ajaydandge.dev/): Online code runner to write, run, and test code - [Contests](https://contests.ajaydandge.dev/): Aggregator for upcoming competitive programming contests ## Writing - [Python Path Resolution](/blog/python-path-resolution.md): How sys.path[0] behaves when running scripts from nested folders, why sibling imports break, and the difference between PYTHONPATH and python -m. - [Precision and Recall](/blog/precision-and-recall.md): A practical mental model for keeping precision and recall straight using a simple spam detection example. - [Benchmarking Local LLMs: Where VRAM Limits, MoE, and PLE Architecture Collide](/blog/local-llm-inference-benchmark-gpu-vs-cpu-only.md): Benchmarking Google Gemma 4 (MoE, dense, PLE) and Mistral Small 3.2 on an NVIDIA GeForce RTX 4060 Ti (16GB VRAM, Intel Core i7-12700F, 32GB RAM) vs an Intel Core i5-13420H (16GB RAM, no discrete GPU) across VRAM offloading, prompt eval, and token generation. - [Optimizing Mobile PageSpeed Insights: How I Reached 99 on Performance](/blog/optimizing-mobile-pagespeed-insights-how-i-reached-99-on-performance.md): A practical breakdown of how deferring analytics scripts, code-splitting client components, and updating transpilation targets boosted this blog's Mobile PageSpeed score from 85 to 99/100. - [Deploying Claude Usage: SQLite, Postgres, or MySQL — and Google Sign-In](/blog/deploying-claude-usage-sqlite-postgres-or-mysql-and-google-sign-in.md): The follow-up to building Claude Usage: making it deployable — splitting a session's history into per-turn timestamps, running on PostgreSQL or MySQL as well as SQLite, and adding Google sign-in alongside Microsoft Entra ID. - [Building Claude Usage: Per-Person Cost Tracking on a Shared Claude Account](/blog/building-claude-usage-per-person-cost-tracking.md): A plugin, a FastAPI and SQLite server, and a dashboard that track per-person token usage and cost when a team shares Claude accounts. - [Using Claude Code with Non-Anthropic Models via LiteLLM](/blog/using-claude-code-with-non-anthropic-models-via-litellm.md): Setting up Claude Code to use non-Anthropic models, including free ones on OpenRouter, by running a local LiteLLM proxy in front of it. - [Burrow Adds a Read-Only Guard and Agent Skills](/blog/burrow-adds-read-only-guard-and-agent-skills.md): Three additions to burrow: per-profile read-only access that blocks write statements at the client level, a non-interactive config set command for scripting, and skill installation across multiple coding agents via the Agent Skills standard. - [Burrow Adds MySQL, Secure Passwords, and a Claude Code Skill](/blog/burrow-adds-mysql-secure-passwords-and-claude-code-skill.md): Burrow now supports MySQL alongside PostgreSQL, stores passwords securely outside config.toml, and ships a proper installer for the Claude Code skill. - [Burrow Now Supports Direct Connections](/blog/burrow-now-supports-direct-connections.md): A small addition to burrow: profiles can now connect to PostgreSQL directly, without an SSH tunnel. - [The OS Knows the Timezone, Not Your Language](/blog/the-os-knows-the-timezone-not-your-language.md): The OS knows the timezone, not the language. Everything else — ambiguous strings, broken queries, wrong date partitions — follows from that. - [How I Understand LLMs](/blog/how-i-understand-llms.md): A practical mental model of how large language models work — from tokens and probabilities to inference, context windows, and why they hallucinate. - [Building Burrow: A CLI for Querying Databases Behind a Bastion](/blog/building-burrow-cli-for-querying-databases-behind-bastion.md): How I built a developer CLI that tunnels through a bastion host over SSH to query a PostgreSQL database — and the small decisions that made it feel right. - [OAuth vs OpenID Connect: Stop Mixing Them Up](/blog/oauth-vs-openid-connect.md): OAuth is about authorization (what an app can do), while OpenID Connect is about authentication (who the user is). - [Building User Interface for Querying ChromaDB](/blog/building-ui-for-querying-chromadb.md): Building a lightweight web interface for exploring and querying ChromaDB collections using Ollama-powered embedding models. - [What CORS Is, What CORS Is Not](/blog/what-cors-is-what-cors-is-not.md): CORS is often misunderstood as a security mechanism, but it's really just a browser behavior filter. This post clears up common myths and explains what CORS actually does—and doesn't do. - [Multi-Account Git Setup with SSH and Commit Signing](/blog/multi-account-git-config.md): A guide to setting up multiple Git accounts (e.g., work and personal) across different hosting providers like GitHub and GitLab using SSH keys, directory-based Git configuration, and commit signing