Amulya Varshney
Verified Expert in Engineering
AI Software Engineer and Developer
Bangalore, India
Toptal member since September 9, 2026
Amulya is an AI software engineer with 3+ years of experience building production AI agents, LLM systems, and distributed back-end platforms at Intuit, Razorpay, and ZEISS. He is experienced in multi-agent orchestration, RAG, tool calling, memory, evaluations, observability, and voice AI, with strong skills in Python, Go, Java, and TypeScript. Amulya owns technically hard AI systems end to end, from architecture and evaluation through production reliability and customer impact.
Portfolio
Experience
- TypeScript - 4 years
- Problem Solving - 4 years
- REST APIs - 4 years
- Python - 4 years
- Next.js - 4 years
- Design Patterns - 4 years
- Java - 3 years
- Retrieval-augmented Generation (RAG) - 3 years
Preferred Environment
Docker, Kubernetes, A2A, Agentic AI, Agentic AI Systems, Agent Memory, AI Agents, AI Prompts, Design Patterns, Model Context Protocol (MCP), Claude Code
The most amazing...
...AI system I've built is a multi-agent orchestration platform for QuickBooks Payroll & Tax that automated complex workflows and improved reliability.
Work Experience
AI Software Engineer II
Intuit
- Designed and shipped a production-grade multi-agent AI system for QuickBooks Payroll and Tax, orchestrating planning, retrieval, reasoning, tool execution, and deterministic fallbacks to automate complex, multi-step customer workflows.
- Architected the core agent execution workflow and orchestration platform with dynamic routing, structured tool calling, retries, guardrails, RAG, and deterministic fallbacks, enabling reliable execution across business-critical AI workflows.
- Integrated QuickBooks Online Payroll and Tax across ChatGPT, Claude, and Perplexity, enabling customers to query payroll data and execute actions through natural-language conversations securely.
- Built scalable state and memory infrastructure for conversation history, cross-session context, contextual retrieval, personalized multi-turn interactions, retries, and failure recovery across distributed AI services.
- Owned production LLM evaluation and observability infrastructure, including prompt/tool tracing, automated regression testing, retrieval and routing metrics, quality evaluation pipelines, and safe iteration across models and prompts on production.
- Improved latency, throughput, and reliability of payroll-critical services through intelligent caching, inference optimization, database tuning, failure recovery, and zero-downtime migrations.
Applied AI Engineer, AI Platform
Razorpay
- Architected and launched Ray, an enterprise AI agent using Python, FastAPI, LangGraph, and LLMs, automating more than 70% of merchant support requests and reducing resolution time from hours to seconds.
- Built and evaluated voice AI workflows integrating Sarvam AI, ElevenLabs, and OpenAI STT/TTS, testing prompt behavior, transcription/synthesis quality, multi-turn conversation flows, and end-to-end latency for production-oriented voice experiences.
- Owned end-to-end agentic workflows spanning planning, retrieval, reasoning, tool execution, refinement, contextual state, error recovery, and production APIs for complex fintech scenarios.
- Designed back-end infrastructure for high-concurrency AI workloads, integrating internal services and knowledge systems while handling retries, contextual state, error recovery, and production reliability.
- Engineered an internal LLM evaluation platform in Go for prompt regression testing, automated quality scoring, safety validation, and continuous benchmarking across prompts and model versions.
- Developed evaluation metrics for factual accuracy, instruction adherence, completeness, and financial compliance, improving measured production AI response quality by 40%.
Software Engineer
Zeiss
- Built and deployed secure, scalable back-end services using Java, Spring Boot, REST APIs, and RBAC for enterprise security products.
- Designed a real-time gRPC-based distributed notification system for threat alerts, system updates, and configuration changes, improving incident response time by 60%.
- Architected full-stack product features using C#, React, and TypeScript, owning back-end APIs, reusable front-end components, and scalable configuration workflows supporting 1,000+ concurrent users.
Experience
VitalPlan, Multimodal AI Nutrition Platform
https://amulyavarshney.github.io/VitalPlanCare Gap Closer, Agentic AI for Preventive Healthcare
Education
Bachelor's Degree in Computer Science and Engineering
International Institute of Information Technology, Naya Raipur (IIIT-NR) - Naya Raipur, India
Certifications
Multi AI Agent Systems with crewAI
DeepLearning.AI
Developing Generative AI Applications on AWS
Amazon Web Services
Skills
Libraries/APIs
REST APIs, React, Playwright, Node.js
Tools
ChatGPT, Claude, AI Prompts, AI SDK, Claude Code
Languages
Python, TypeScript, JavaScript, SQL, Java, C++, Go, C#, GraphQL
Frameworks
LangGraph, Next.js, Spring Boot, gRPC
Paradigms
REST, Microservices, Model Context Protocol (MCP), Design Patterns, Test Automation, Role-based Access Control (RBAC)
Platforms
Langfuse, Docker, Kubernetes, Azure
Storage
PostgreSQL, MySQL, Redis, MongoDB, Elasticsearch
Other
Perplexity, Retrieval-augmented Generation (RAG), FastAPI, Large Language Models (LLMs), Multi-agent Systems, A2A, Tool Calling, LLM Evaluation, LangChain, Multimodal GenAI, Multimodal Models, Software Engineering, Problem Solving, Agentic AI, AI Agents, Agentic AI Systems, RAG Systems, LLM Agents, Artificial Intelligence (AI), Agentic Coding, Platform Engineering, ElevenLabs, Kafka, A2UI, Agent Memory, Large Language Model Operations (LLMOps), Embeddings from Language Models (ELMo), GCP, CI/CD Pipelines, Temporal, Amazon Bedrock AgentCore
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring