Why AI Coding Agents Need Git-Native Regression Guardrails
How LLM context limits and stateless prompt loops cause agents to repeat previously solved mistakes, and how deterministic git-level memory fixes this.
Hy, I'm Pavan
Software Builder & Product Engineer with a founder mindset.
I'm a developer who understands how software products and businesses scale, not just the code. 4+ years engineering software across frontend, backend, developer tooling, and autonomous AI agents. Hand me a hard problem or a blank slate — I'll architect the plan, build it with precision, and ship it. I adapt, learn, and deliver.
Top Products
Spatial tools, git-native agent memory, and developer infrastructure
Things I've built
Developer tools, spatial canvases, and products.
Sticky Note Wall — Spatial Canvas
Low-latency infinite spatial canvas for real-time collaborative note-taking, brainstorming, and multi-user feedback. Built with instant sync and lightweight client rendering.
scars — Git-Native Regression Guardrails
Git-native regression guardrails and failure memory for AI coding agents and engineering teams. Deterministic recall, token-budgeted prompt injection, and Model Context Protocol (MCP) server.
Neotic — Contextual Release Delivery
Contextual in-app changelog and release delivery designed for modern web applications, SaaS products, and agentic workflows.
BuiltWhileBroke — Engineering Hub
Independent engineering hub documenting production architecture blueprints, systems design, and lessons learned while building and scaling software.
Get in Touch
Have an interesting project, systems challenge, or developer tooling problem to solve? Let's connect.
Blog
Thoughts on software architecture, agentic AI workflows, and building products.
Engineering a Low-Latency Spatial Canvas with TypeScript
Architecting Sticky Note Wall: virtualized canvas transformations, real-time WebSocket broadcasting, and responsive layout coordination.
Building Production Model Context Protocol (MCP) Servers in Go
A practical guide to implementing JSON-RPC 2.0 transport, tool definitions, and token-conscious context retrieval for agent ecosystems.
Autonomous AI Agents: Benchmarking Real-World Execution Loops
An objective engineering analysis of autonomous agent frameworks: reliability, error-correction loops, token expenditure, and production readiness.
Context-Aware In-App Release Delivery for Modern Web Apps
Why traditional email changelogs get ignored, and how in-app contextual SDKs increase user engagement and adoption of new product capabilities.
Local-First Architecture for High-Performance Developer Tools
Exploring SQLite in-process storage, zero-cloud dependencies, and instantaneous startup times for terminal and editor tooling.