I build the agents.
They build the software.

Seven exits. Twenty years of enterprise architecture. Now engineering autonomous systems that think, ship, and self-correct.

Arun Prasath E G
20+ years · 7 exits · 50+ projects
5,599 contributions this year
26 weeks of contributions 5,599 this year
About

Seven exits. One playbook:
architecture that endures.

Anyone can generate code today. A teenager with a laptop can spin up a Next.js app with AI. What they can't do is architect systems that survive production — enforce security at the boundary layer, design feedback loops that catch regressions before they ship, or make the decisions that separate a prototype from an enterprise product.

Every company I've built followed the same model — lean teams with total architectural ownership. My autonomous development systems aren't "AI writes code fast." They're engineered with test gates, static analysis, backpressure mechanisms, and quality validation at every stage. The AI handles volume; 20 years of enterprise experience ensures every line that ships is production-worthy.

I build multiple products in parallel, each with a clear path to revenue. But speed without discipline is just technical debt at scale. The architecture is the differentiator — feedback loops that self-correct, security-first design, and the human judgment that comes from two decades of knowing what breaks at 3am.

“Code generation is a commodity. Architecture is not. The 20-year advantage isn't typing speed — it's knowing what to build and where it will break.”
— The operating model
Core Focus

AI agents are not a feature I add.
They're how I build everything.

From autonomous coding loops to self-improving security auditors — agent architecture is the thread connecting everything I ship.

01
Agent Orchestration Frameworks
Desktop apps and CLI tooling that turn LLMs into autonomous software engineers. Planning mode, build mode, tree-sitter AST analysis, loop engines with automatic commit/test/iterate cycles. Full orchestration with crash recovery, backpressure, and token tracking.
02
Self-Improving Audit Pipelines
GEPA (Genetic-Pareto) text evolution engines that optimize investigation strategies through a three-stage pipeline: execute, reflect on failure traces, curate improved versions via Pareto-efficient search. Static analysis, graph-based code exploration, LLM-as-judge evaluation. Reflection over gradients, not brute-force optimization.
03
Multi-Agent Architectures
Production systems where a primary agent spawns parallel sub-agents for concurrent research, synthesizing structured findings into unified analysis. Claude Agent SDK, MCP servers, tool-use pipelines, real-time WebSocket streaming.
04
AI Safety Infrastructure
Production microservices for prompt injection detection, hallucination validation, stuck-loop detection (repeater, oscillator, wanderer patterns). Pay-per-call with on-chain micropayments.
05
Small Model Selection & Fine-Tuning
Anyone can throw a problem at Opus or Gemini Pro. The real skill is knowing when a fine-tuned 3B model outperforms a 400B one — at 1/100th the cost and 50x the speed. Domain-specific fine-tuning, evaluation pipelines, and production AI at scale without burning margins on API calls.
06
DSPy & Conversational AI
Learnable graph architectures with goal-oriented conversation management, engagement detection, structured data extraction through natural dialogue. Optimizable modules with MLflow tracking.
Systems Engineering

Beyond agents —
the infrastructure underneath.

Rust · WebSockets · SOCKS5

Low-Latency Execution

Sub-10ms order routing through proxy-optimized tunnels. O(1) trigger evaluation engines. Lock-free order book reconstruction. 15x latency improvements from rewriting proxy layers.

Solidity · Flashbots · MEV

On-Chain Protocols

Flash loan architectures, MEV bundle construction, Dutch auction mechanics, builder economics optimization. Execute or discard within a single block. Zero capital at risk.

TypeScript · Next.js · Stripe

Full-Stack Products

Multi-tenant SaaS platforms end-to-end. Billing automation, IaC deployments, AI-integrated analytics, e-commerce engines. From Ansible playbooks to AI chatbots.

Code generation is a commodity.
Architecture is the moat.
Technical Stack

The tools behind
the velocity.

AI & Agents
Claude CodeClaude Agent SDK MCP ServersDSPy 3.0 Pydantic AIGEPA MLflowTree-sitter
Performance
RustWebSockets SOCKS5 TunnelingLock-Free Structures TauriSolidity FlashbotsFlash Loans
Web & SaaS
TypeScriptNext.js TanStackMedusa Stripe
Data & Infra
PostgreSQLRedis ClickHouseDuckDB DockerCoolify RailwayHetzner GCPAnsible
Contact

Let's build
something ambitious.

Open to high-impact roles where AI-native engineering meets real business outcomes. If you're building something that matters, I want to hear about it.