Headline

AI Solutions Architect (Business Analyst) @ Amazon | Building Production GenAI Systems at Amazon | Bedrock · Multimodal AI · Step Functions · HITL

About

I build production AI systems - not prototypes, not demos. Systems that run at scale, get measured, and hold up under real operational load. I'm a Business Analyst at Amazon by title, with 9+ years in data and AI - including 6+ years at Amazon, where I've moved from data analytics into GenAI/LLM solution architecture. I design systems that combine large language models with event-driven AWS infrastructure to solve problems that used to require armies of manual reviewers. ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ 🔹 What I've architected: → TRIDENT :- A multimodal GenAI classification system (Amazon Bedrock — text + image reasoning) that automates compliance attribute labelling. Cut per-attribute review time from 217 seconds to 16 seconds at a 5.5% human-override rate — scaling across 450 product categories and 25 global marketplaces, delivering ~$3.8M+/year in savings at full scale. → TRON :- A companion AI meta-authoring system that automates the prompt-engineering process itself, where one LLM writes the classification rules another LLM follows. Projected to cut prompt-authoring time by ~97% and save ~$1.1M across the same 450 product types. → OATH :- A serverless customs data transfer pipeline delivering duty-accurate EU import declarations across 8 languages for 2 major customs brokers across the EU-27, processing 33,500+ records daily. → EU ASIN Compliance Dashboard :- Real-time regulatory risk monitoring for EU customs compliance, and analytics platforms that have informed $50M+ in strategic decisions. ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ 🔹 How I work: I care about the boring parts of architecture as much as the exciting parts — trade-off analysis, build-vs-buy decisions, integration design, reconciliation logic. The AI system is only as good as the pipeline, auth model, and human-in-the-loop review process around it. ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ 🔹 Stack: AWS (Lambda, Step Functions, CDK, Athena, S3, EventBridge, DynamoDB, IAM) · Amazon Bedrock · Redshift · PostgreSQL · Python · PySpark · SQL ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ Always happy to connect with people working in this space that keeps on evolving.

Location

Bengaluru

AI stack

claude code

hugging face

kiro

langchain

langgraph

Work experience

AI Solutions Architect (Business Analyst) at Amazon, Sep 2022 - Present · 4 yrs 1 mo, Bengaluru · Hybrid, Architecting production GenAI systems for trade compliance - from multimodal AI pipelines to self-improving prompt authoring platforms. ▸ TRIDENT - Architected and deployed a production multimodal GenAI labelling system (Amazon Bedrock Claude Sonnet, text + image reasoning) with AWS Step Functions orchestration (8 Lambda functions), processing ~990K ASINs across 450 product types and 25 global marketplaces. Reduced review time from 217 to 16 seconds per attribute with only 5.5% human override rate, delivering ~$3.8M+/year in operational savings. ▸ TRON - Designed a companion AI meta-authoring system that reverse-engineers human-verified labels into production-ready classification prompts. The LLM writes the rules another LLM follows. 7-stage Step Functions pipeline with self-improving auto-evaluation loops. Projected ~97% time reduction and ~$1.1M cost savings across 450 PTs. ▸ OATH - Designed and shipped a serverless integration architecture delivering classification data to global customs brokers (UPS, Maersk) across the EU-27. GRASP Proxy (Graph API) → S3 → Lambda → Redshift pipeline with OAuth 2.0 + SigV4 dual-identity authentication, processing 33,500+ records daily across 8 business lines. ▸ Designed a real-time regulatory risk monitoring architecture for EU customs compliance with AI-driven predictive risk-scoring across 5 cross-border trade programs. ▸ Designed a Bedrock-powered agentic reporting architecture with anomaly detection and natural-language variance analysis. ▸ Engineered unified analytics across 80+ metric dashboards and 4 disparate systems, saving 234+ hours/year. ▸ Designed large-scale ETL pipeline architecture processing 5TB+ of daily data for 1,000+ users org-wide. Skills: Generative AI · Amazon Bedrock · AWS Step Functions · AWS Lambda · Solutions Architecture · Python · Prompt Engineering · Serverless Architecture

Workflow Analyst → Senior Associate at Amazon, Dec 2019 - Aug 2022 · 2 yrs 9 mos, Bengaluru · Hybrid, Built the data and analytics foundation that evolved into the current GenAI architecture practice - ETL pipelines, operational dashboards, and cross-functional reporting systems. • Workflow Analyst (Apr 2020 – Aug 2022) • Senior Associate (Dec 2019 – Apr 2020) Skills: SQL · Python · Redshift · ETL Pipeline Design · Data Analysis · Quicksight

Area Business Manager (Data Analyst) at OYO, May 2018 - Dec 2019 · 1 yr 8 mos, Goa, India, Key Focus: 5+ ETL pipelines from scratch, transitioning teams off Excel to interactive Power BI dashboards, and facilitating strategic cost savings.

Data Analyst at RIVIGO, Jul 2017 - May 2018 · 11 mos, Mumbai, Maharashtra, India

Education

Master's degree, Data Science, Deakin University

Post Graduate Program, Data Science and Business Analytics, Texas McCombs School of Business