Hey — thanks for dropping by 👋

Backend-first. Problem solver. Building intelligent systems, not just calling APIs.

I built my foundation in Node.js and backend engineering — REST APIs, auth, databases, the unglamorous stuff that has to actually work in production. Give me a messy repo, an ambiguous bug, or unclear requirements, and I'll break it down and ship something clean.

Right now I'm extending that foundation into AI-powered systems: multi-agent LLM pipelines, observability for agent execution, and backend architecture that treats "AI feature" as a real production concern instead of a demo. Both projects below came out of that — built to solve problems I actually have, not to pad a portfolio.

Open-source is where most of my real learning happened, and I try to give back where I can.

• 🔎 Looking for: Node.js Developer | Backend Engineer | Backend + AI Engineer roles
• 📚 Deepening: LLM orchestration, RAG pipelines, applied ML for backend services

Multi-agent pipeline (resume parsing → JD analysis → gap detection → question generation → evaluation → coaching) orchestrating Gemini, OpenAI, and Anthropic SDKs behind an Express + TypeScript backend. Full OpenTelemetry tracing across every agent span, Supabase Postgres with Row-Level Security, JWT + 2FA/TOTP auth.

Learning platform where finishing curriculum and getting project submissions reviewed actually pays out — via Stripe Connect with manual fallback for unsupported regions. React 19 + Express + Supabase, JWT + Google OAuth, Zod-validated request schemas throughout. Actively in development.

Backend Engineering

• REST API design with clean, layered architecture (controller-service-DTO)
• Auth: JWT, session strategies, RBAC
• Database design: PostgreSQL (Supabase, Row-Level Security), MongoDB, MySQL
• Caching & async processing: Redis, BullMQ
• Error handling and logging that's actually useful when something breaks
• CI/CD, Dockerized deployments

AI / Intelligent Systems

• Multi-agent LLM pipelines (Gemini, OpenAI, Anthropic SDKs)
• OpenTelemetry instrumentation for AI agent execution
• RAG pipeline experimentation
• Prompt engineering with an evaluation mindset, not vibes
• Integrating AI as a real backend feature, not a standalone demo

1. Understand the actual business impact first.
2. Break the problem into the smallest testable pieces.
3. Build incrementally.
4. Add logging and monitoring early, not after something breaks.
5. Refactor once the shape of the problem is actually clear.

I'd rather ship something simple and observable than something clever and fragile.

Reading real production code taught me more about tradeoffs and system structure than any course did. I try to contribute where I can and keep learning in public.

• LinkedIn: https://www.linkedin.com/in/devanshukoli/
• Blog: https://dev.to/devanshukoli/
• Twitter/X: https://twitter.com/Devanshukoli

If you're building something meaningful with AI in production — not just a demo — let's talk.