AI Systems Builder | Autonomous Agents | Developer Tooling Building practical, reliable AI systems that can do more than look impressive in a demo.

I'm an AI-focused builder working on autonomous agents, computer-use automation, and the infrastructure that keeps ambitious systems dependable once they meet real tools, real users, and real-world constraints. My work spans multi-agent orchestration, sandboxed coding environments, LLM reliability, and developer products.

I care about useful systems, thoughtful product design, and safeguards that keep a swarm of agents from becoming a very confident expense report.

• Autonomous and multi-agent systems for planning, coding, testing, and review
• Agent reliability: budgets, backpressure, retries, circuit breakers, and observability
• Sandboxed developer tooling and Docker-based execution environments
• Computer-use automation and systems integrations
• AI product engineering with Rust, Python, TypeScript, and modern web tooling

Languages

AI, Agents & Reliability

🧠 Multi-agent orchestration · LLM evaluation · Agent memory · Computer use · Observability

Cloud & Developer Infrastructure

Niki — Hermetic Multi-Agent Coding System

• A Rust-based system where isolated agents plan, code, test, and review inside Docker sandboxes
• Produces reviewable Git branches instead of giving every agent the keys to the production kingdom

Backstop — AI SDK Reliability Layer

• In-process controls for backpressure, budgets, retries, circuit breaking, and metrics
• Designed to put sensible brakes on AI SDK workflows before cost, rate limits, or chaos take the wheel
• Website and documentation

Phantom — Background Computer-Use Agent

• A Rust-powered Windows automation agent built to work quietly in the background

policyctl — policyctl provides a deterministic policy runtime that sits between the agent and your codebase.

• Provider-agnostic deterministic policy runtime for coding agents.

• Building agent systems with reliability and containment as first-class features
• Shipping tools across Rust, Python, TypeScript, Docker, and LLM APIs
• Exploring the bridge between agent research and practical developer workflows
• Focused on products that make capable AI systems safer, clearer, and more useful

• 🚀 Website: Backstop — hard limits for AI agents
• 💼 LinkedIn: Ravani Roshan
• 🌐 Portfolio: roshans-portfolio-site.vercel.app
• 🧪 GitHub: RavaniRoshan
• 🐦 X: @RoshanAIs
• 📧 Email: [email protected]

agent status: awake
coffee level: suspiciously high
build target: useful systems, not shiny demos