Anoshor Paul
Software Engineer
▌
I build the plumbing that makes AI agents actually useful. At RingCentral I work on an enterprise agentic AI assistant platform — backend microservices, Model Context Protocol (MCP) integrations, and developer SDKs that let AI agents discover, configure, and execute enterprise skills.
I joined as one of the first ~10 engineers on RingCentral's first dedicated AI engineering team in India. Off the clock I ship terminal games, build backends for real clients, and grind LeetCode (2175 · top 2%).
// click a branch to expand it
- One of the first ~10 engineers on RingCentral's first dedicated AI engineering team in India.
- Designing backend microservices for skill execution, configuration, recommendation, and orchestration.
- Shipped MCP-based skill integrations powering 30+ customizable enterprise agents, with MCP and A2A interoperability.
- Co-building the agent Skill SDK (CLI) — adopted by 10+ teams to author and deploy their own agent actions.
- Shipped React.js frontend features and fixed production issues in Java backend services.
- Built an internal engineering dashboard from scratch — now adopted company-wide for engineering stats and AI usage analytics (tokens, credit spend, per-team AI budgets).
- Prompt design and evaluation for LLM-based action-execution response reasoning.
TypeFaster
Terminal-first typing game — live multiplayer races, ghost replays, daily challenges, and a per-key typing coach. Published on PyPI & Homebrew.
github ↗Building Management System
Microservices backend for a property-management product shipped to a US client — JWT auth, leases & maintenance, Stripe payments (cards, ACH, webhooks), S3 + CloudFront.
github ↗GeoGuide Bot
Vision-guided robot for the E-Yantra Robotics Competition 2023-24 — ranked top 30 globally.
github ↗Resilient Kannada Scene Text Detection: CRAFT-YOLOv8 Fusion
2024 11th International Conference on Computing for Sustainable Global Development (INDIACom) · IEEE · 2024
A fused CRAFT + YOLOv8 pipeline for detecting Kannada text in natural scene images, addressing the lack of scene-text detection models for the language. We built a custom annotated dataset, automated the annotation pipeline to remove the manual labelling bottleneck, and improved detection accuracy by 10% over the prior baseline.
Building something in the agentic AI or backend space, or hiring for one? My inbox is open.