Harsh Soni

Software Engineer specializing in scalable systems and high-performance applications. Over 3 years of production experience at Attentive.ai and Infosys shipping solutions for 500+ enterprise clients. Currently pursuing an MS in Computer Science at the University of Florida, with strong expertise in Python, React, FastAPI, and production LLM/RAG pipelines.

Education

University of Florida

Masters in Computer Science • GPA: 3.8

Connection

LinkedIn open_in_new GitHub open_in_new

Experience

Software Development Engineer

Attentive.ai • Noida, India • Jun 2023 — Jun 2025
  • Diagnosed a critical 12s settings-page load bottleneck impacting 500+ enterprise clients; independently owned and shipped a layered fix using React code-splitting, debounced zoning calculations, and IndexedDB caching, reducing load time by 86% (to 1.7s) and directly reducing client churn risk.
  • Identified slow construction document Q&A as a top client pain point and independently built Beam GPT end-to-end: an LLM-powered assistant using RAG (Pinecone vector DB + OpenAI GPT-4) that cut task completion time by 60% for 500+ users.

Software Engineer (Specialist Programmer)

Infosys • Bangalore, India • Feb 2022 — Jun 2023
  • Migrated a CLI-based internal tool to a React/Flask web app for Apple IS&T, replacing manual workflows with a browser UI backed by Redis caching to reduce repeated query latency.
  • Contributed 5 merged PRs to neo4j-graphql-java resolving race conditions in the query executor and hardening GraphQL schema validation, improving concurrency correctness across 3 internal teams.

Projects

Automated Multi-Agent PR Reviewer (PR Sentinel)

A production-deployed GitHub App on GCP Cloud Run that automates PR reviews with zero human intervention. Architected a multi-agent RAG pipeline using LangGraph that spawns parallel sub-agents (Security, Performance, and Code Quality) with autonomous tool-calling loops over Pinecone vector indexes. Engineered with Celery/Redis for durable queue processing, HMACS-SHA256 signature verification, and namespace-scoped incremental vector ingestion reducing ingestion time by ~90%.

FastAPI LangGraph Celery Redis Pinecone GCP Cloud Run
View on GitHub arrow_forward

AI-Powered Histopathology Scoring System

A diagnostic tool for pancreatic tissue analysis built by fine-tuning ResNet18 (transfer learning on ImageNet weights) to achieve 4-metric pathology scoring, reducing manual analysis time by 40%. Engineered a client-side upload queue with a FIFO buffer to handle 50MB+ TIFF batches, preventing server RAM exhaustion by throttling concurrent FastAPI requests via async/await.

Next.js FastAPI PyTorch ResNet18 PostgreSQL Docker
View on GitHub arrow_forward Live Demo arrow_forward

Writing

View All Blog Posts & Research Summaries arrow_forward

Skills

Languages

Python, TypeScript, JavaScript (ES6+), C++, Java, SQL

Frameworks & Libraries

React, Next.js, Redux, FastAPI, Flask, Node.js, PyTorch, LangGraph, LangChain, Tailwind CSS

Tools & Platforms

Docker, Git, PostgreSQL, MongoDB, Redis, Pinecone, GCP Cloud Run, AWS, GitHub Actions, Linux

Get In Touch

Let's architect something extraordinary.