Usharani Choudhury

CTO

Bengaluru, Karnataka, India12 yrs 10 mos experience

Most Likely To SwitchAI ML Practitioner

Key Highlights

Built GenAI platforms with 99.5% uptime.
Reduced deployment time by 92% through automation.
Optimized AWS costs by 30% while scaling.

Stackforce AI infers this person is a SaaS expert specializing in Agentic AI and Generative AI systems.

Contact

Skills

Core Skills

Agentic Ai DevelopmentGenerative AiPython

Other Skills

FastAPIGENAILangGraphRAGAWSHybrid RetrievalPython (Programming Language)Model Context Protocol (MCP)Vector DatabasesAmazon BedrockAI AgentsdockerSQLGitlabFlask

About

Most GenAI systems fail in production due to poor control, weak retrieval, or unbounded agents. I build systems that don’t. I build production-grade Agentic AI and GenAI systems that don’t break under real-world conditions. Specialized in multi-agent systems, RAG pipelines, and LLM-powered workflows, with a strong focus on failure handling (agent loops, hallucinations, tool misuse) and production reliability. In my recent work, I’ve: Built and scaled GenAI platforms serving 500+ enterprise users with 99.5% uptime. Designed multi-agent systems (LangGraph) orchestrating planning → retrieval → reasoning → execution Improved RAG accuracy to 90%+ using hybrid retrieval and evaluation-driven optimization Reduced deployment time by 92% (2 days → 4 hours) through CI/CD and platform automation. Optimized cloud infrastructure, reducing AWS costs by 30% while supporting 10,000+ daily API calls I’ve worked extensively with OpenAI and Amazon Bedrock, building systems with tool-calling, retrieval optimization, and evaluation frameworks for real-world enterprise use cases.

Experience

12 yrs 10 mos

Total Experience

4 yrs 3 mos

Average Tenure

6 yrs 7 mos

Current Experience

Infosys

2 roles

Technology Lead Infosys

Nov 2019 – Present · 6 yrs 7 mos · Bangalore Urban, Karnataka, India · Hybrid

Building production-grade agentic systems using LangGraph, focused on reliability, observability, and failure handling (loops, hallucinations, tool misuse).
Designed multi-agent workflows with HITL, evaluation pipelines, and real-time tool integration across enterprise systems.
Systems deployed to 500+ users with 99.5% uptime, supporting large-scale document intelligence and decision automation.
Designed and deployed multi-agent orchestration workflows across planning → retrieval → reasoning → tool execution → validation
Built secure tool-calling architecture with schema validation, retries, and fallback routing for production reliability
Diagnosed and mitigated agent failure modes including infinite loops, tool misuse, context drift, and hallucinated outputs
Implemented human-in-the-loop (HITL) checkpoints for low-confidence decisions and escalation handling
Built evaluation pipelines using golden datasets to measure retrieval quality, grounding accuracy, and agent behavior
Developed observability pipelines capturing reasoning traces, tool decisions, and execution paths
Delivered RAG pipeline with hybrid retrieval (vector + metadata + reranking), improving accuracy to 90%+
Built scalable APIs using FastAPI supporting real-time AI workflows and integrations
Reduced AWS cost by 30% and improved deployment efficiency by 92% using CI/CD automation