Looking for a job

Harsh Kumar

@harshkumar41

Backend engineer and founder at Patience AI, building real-time AI voice agents and products serving 500+ concurrent LLM queries.

India

Message

At Patience AI, I founded and operate the platform end to end, working across its backend, frontend, infrastructure, and releases. I built a real-time AI voice agent that answers website calls in Hindi and English, books meetings without double-booking, and can bring a human into the call.

I launched KitabPadhoIndia, a campus marketplace with 200+ REST endpoints, a WebSocket chat gateway, and more than 4,500 automated tests. Its Flutter Android app and PWA help verified college students buy, sell, and rent used books.

At Centre for Development of Advanced Computing, I served 500+ concurrent LLM inference queries at sub-2s P95 with asynchronous FastAPI endpoints over HPC infrastructure. At Fusionstak LLC, I reduced average API response time from 1.2s to 0.4s using Redis caching and SQL query optimization.

Experience

Work history, roles, and key accomplishments

Current

Founder & Lead Engineer

Current

Patience AI

Founded and operate the platform end to end with 59 REST endpoints and 4 role-scoped portals. Built a real-time AI voice agent and a database-backed CMS, serving as sole engineer across backend, frontend, infrastructure, and releases.

Python fastAPI LiveKit Claude APIs GROQ Whisper PostgreSQL Redis Cloudflare GitHub Actions Nginx Docker

Associate Software Engineer

Fusionstak LLC

Nov 2024 - Mar 2025 (4 months)

Reduced average API response time from 1.2s to 0.4s and eliminated 60% of redundant requests by introducing Redis caching and optimizing SQL queries. Reduced third-party API latency by 35% through request batching, exponential backoff, and circuit-breaker patterns.

Redis SQL .NET 6.0 EF Core Formly JSON API Optimization Caching Circuit Breakers

Software Development Intern

Ministry of Housing and Urban Affairs

Apr 2023 - Jun 2023 (2 months)

Increased daily active users by 20% through analytics-driven product improvements. Project selected among top 5 of 40+ projects for presentation to senior officers.

Analytics Product Improvement User Engagement

Software Development Intern

Centre for Development of Advanced Computing (CDAC)

Served 500+ concurrent LLM inference queries at sub-2s P95 by building asynchronous FastAPI endpoints over HPC infrastructure. Improved RAG retrieval precision by approximately 30% by redesigning document chunking and enriching metadata during ingestion.

fastAPI HPC RAG BLEU ROUGE LLM Python