Adeppa Thondur

Bengaluru

Adeppa Thondur on Tal

Headline

About

Principal-level Backend & GenAI Systems Architect with 15+ years designing and scaling enterprise-grade distributed systems, agentic AI platforms, and enterprise search solutions. Currently leading the enterprise GenAI transformation of a 70M+ annual-call IVR ecosystem at Lowe's — migrating legacy NLU to a Google ADK-based multi-agent architecture using Agent2Agent (A2A) for task delegation and Model Context Protocol (MCP) for standardized tool/API integration, increasing call containment from 18% to 40% while cutting LLM token and inference costs by 50%. Architected production RAG pipelines using OpenAI GPT-4/5, Whisper, and FAISS for real-time transcript intelligence and contextual retrieval at 100K+ calls/day, with full OpenTelemetry instrumentation and Arize-powered evaluation across the orchestrator, subagents, and retrieval layer. Built domain-driven microservices (FastAPI, Kubernetes, event-driven Kafka pipelines) doubling platform throughput while maintaining sub-second latency, and re-architected Elasticsearch clusters to cut query latency 40% and accelerate indexing 96%. I operate at the intersection of AI strategy, distributed systems architecture, and engineering leadership — leading and mentoring a 10-member team while partnering with Product leadership to align platform roadmap with enterprise AI strategy.

Location

Bengaluru

AI stack

claude code

langgraph

mistral

openai

windsurf

Google ADK

Work experience

Lead Software Engineer at Lowe's Companies, Inc., Feb 2020 - Present · 6 yrs 7 mos, Bengaluru, Karnataka, India, • Led the enterprise migration of a high-volume IVR ecosystem from legacy NLU to a production GenAI multi-agent platform built with Google Agent Development Kit (ADK), increasing call containment from 18% to 40% across 70M+ annual calls • Architected domain-specialized agents using Agent2Agent (A2A) for task delegation and Model Context Protocol (MCP) for standardized integration with enterprise tools and APIs, with context compaction reducing LLM token cost by 50% • Instrumented production agent observability using OpenTelemetry and Arize across the orchestrator, subagents, LLM calls, tools, and retrieval to accelerate debugging and regression testing • Delivered GenAI IVR analytics using Azure Communication Services, Whisper, OpenAI APIs, RAG, and FAISS, processing 100K+ calls/day for transcription and intent intelligence • Defined cloud-native architecture using FastAPI, Kubernetes, and API Gateway, doubling throughput while maintaining sub-second response times • Built Spring Boot and Apache Kafka-based event-driven microservices for real-time inventory updates, syncing product availability with Elasticsearch and integrating Twilio for outbound customer notifications • Optimized Elasticsearch cluster topology and PySpark ingestion pipelines, cutting query latency 40% and accelerating indexing 96% • Reduced infrastructure cost 50% through platform consolidation and autoscaling across AI and search clusters • Led and mentored a 10-member engineering team through design reviews, technical direction, hiring, and roadmap delivery

Lead Software Engineer at Lowe's Services India Pvt. Ltd., Feb 2020 - Present · 6 yrs 7 mos, Led the enterprise migration of a high-volume IVR ecosystem from legacy NLU and deterministic dialog flows to a production GenAI multi-agent platform built with Google Agent Development Kit (ADK), increasing call containment from 18% to 40% across 70M+ annual calls and reducing live-agent dependency. Architected domain-specialized agents for order status and product availability using Agent2Agent (A2A) for task delegation and Model Context Protocol (MCP) for standardized tool/API integration, with shared session and context continuity across the orchestrator and subagents via controlled state propagation, selective context handoff, and context compaction with rolling summaries; reduced LLM token usage and associated cost by 50% versus the initial monolithic agent architecture. Implemented production agent observability and evaluation by instrumenting OpenTelemetry traces and analyzing them in Arize across the orchestrator, subagents, LLM calls, tools, retrieval, and routing; tracked latency, token usage, failures, and quality signals to accelerate debugging and regression testing. Architected and delivered GenAI IVR analytics and conversation-intelligence capabilities using Azure Communication Services, Whisper, OpenAI APIs, RAG, and FAISS, processing 100K+ calls per day for transcription, contextual retrieval, and intent intelligence. Defined the cloud-native architecture for an internal customer-service platform using domain-driven microservices, FastAPI, Kubernetes, and API Gateway, improving resiliency and doubling throughput while maintaining sub-second response times. Built Spring Boot and Apache Kafka-based event-driven microservices to process real-time inventory updates, synchronizing product availability with Elasticsearch search indexes and integrating Twilio as the SMS aggregator for outbound customer notifications, cutting inventory-to-search sync latency and eliminating manual reconciliation.

Lead Software Engineer at TiLa (FP Tech Science Pvt. Ltd.), Oct 2019 - Feb 2020 · 5 mos, Bengaluru, Karnataka, India

Software Engineer Technical Lead at TiLa (FP Tech Science Pvt. Ltd.), Oct 2019 - Feb 2020 · 5 mos, Implemented e-commerce search capabilities within a Java and Spring Boot microservices architecture for online retail. Built multi-level product documents for fashion and electrical domains and used FastText-based machine learning to predict category names.

Senior Software Engineer at Zilingo, Apr 2019 - Oct 2019 · 7 mos, Greater Bengaluru Area, Setting up an ElasticSearch 6.5 cluster by configuring 3 Master, 5 Data 2 inject node and 1 Client Nodes split across different AWS regions for high availability. Autosuggestion using the phrase Suggester including user feedback calculation based on clicks and views Search functionality is ruining in the micro service architecture using Scala and Play Framework Consuming kafka service for the indexing and primary source of index PostgreSQL Configure Cluster monitor like Grafana and Prometheus, graylog Deep understanding of information retrieval concepts, especially using tools such as ElasticSearch Integrated Scala and Play Framework and Restful Controllers communicating to a custom Elastic client Coded Scala based ES Queries and custom ES Query aggregations using the Transport client API for searching and accessing data in Elastic Search Coded custom data mappings and data analyzers for multiple elastic indices Translating business requirements into an Elastic Search implementation Providing advice regarding analyzers, tokenizes and Elastic Search mappings Determining what Elastic Search queries produce the best search experience Improving performance and scalability of an existing Elastic Search app Ranking the top-k documents using LTR (Learn to Rank), based on the Data Scientist provided models and integrated into Elastic Search Identifying the Named entities using NLP and building query based on Rules, it’s will help better way tune the relevancy

Senior Software Engineer at Decision Resources Group, Apr 2016 - Apr 2019 · 3 yrs 1 mo, Bengaluru, Karnataka, India, Proposed and designed enhancements for the existing Search Module Implemented the technical POCs and presented the same for the Technical and Business stakeholders for approvals. Handling a team of 3 technical developers in Insights Platform project. Playing a role of a technical point of contact for Search Module and related functionalities. Playing a role of a devops engineer for the team. Part of the Insights Platform design committee for any product search related implementations. Part of the Functional and Technical Review committee responsible for reviewing the technical designs and code. Configured 8 node SolrCloud cluster with 3 Zookeeper ensemble and Handling around 90 GB of data from various products ,This data split into 4 shards and replication factor 2 with query response time max 4 sec. Building Recommendation, Clustering and Q&A system using NLP, apache mahout on top of solr. Writing custom request handlers using (NLP with custom entity recognition model) and filter on top of solr sources code. Relevancy tuning using combination learn to Rank (LTR) and SVM model form Click stream data. Building various data pipelines using Apache Spark and Scala on Databricks platform

Education

Master of Computer Applications, Computer Programming, Specific Applications, Osmania University, Hyderabad

Bachelor's degree, Electronics, Sri Krishnadevaraya University

Adeppa Thondur

10+ years of work experience

Adeppa’s AI Stack

Education