Harsh Kumar profile photo 1 on Tal

Harsh Kumar

Senior Software Engineer at Tellius | Building RAG & LLM-Orchestration Systems (LangGraph, Qdrant, Multi-Agent Retrieval)

Greater Bengaluru Area

7+ years of work experience

Tellius
Aug 20251 yr 2 mosPresent

Bengaluru · Hybrid

• Architected and independently built tellius-health-agent, an autonomous FastAPI service that auto-discovers a deployment's Business Views, unstructured workspaces, and workflows, and runs automated + LLM-judge evaluations across 5 capability areas (Text-to-SQL, RAG, routing, MCP, regression) — cutting per-deployment QA time from hours to minutes and shrinking failure-detection latency from customer-reported (days) to a nightly automated cycle. • Authored the technical design and phased implementation roadmap for company-wide AI-quality monitoring spanning structured, unstructured, routing, and MCP testing areas. • Architected and shipped a production RAG platform across 5 enterprise document sources (Google Drive, S3, GCS, Azure Blob, SharePoint) by leading a cross-timezone team of 3 engineers to deliver custom Airbyte connectors with Mistral OCR, Airflow-orchestrated hybrid Qdrant indexing (BM25 + SPLADE + Dense), and a LangGraph multi-agent retrieval service with RRF fusion and cross-encoder reranking.
Tellius
Feb 20223 yrs 6 mosJul 2025

Bengaluru · Hybrid

• Reduced query processing time for moving window calculations by 30% by redesigning the aggregation architecture to support moving average, moving min, and moving sum. • Enabled multilevel sorting and top-N filtering for pivot tables, reducing report generation stepsby 80%, by architecting an extensible pivot table engine. • Improved operational visibility for 5-6 background jobs by designing a real-time task tracking system, reducing mean debugging time by 50%.
Lam Research
Feb 20211 yr 1 moFeb 2022

Bangalore Urban

• Reduced mean time to resolve deadlocks by 90% by implementing automated deadlock detection that captures detailed logs and tool state on trigger. • Improved team technical proficiency across 3 technology domains by leading regular knowledge-sharing sessions, driving continuous learning within the team. • Improved tool throughput by resolving critical escalations and performance bottlenecks identified through systematic profiling.
Lam Research
Aug 20191 yr 7 mosFeb 2021

Bengaluru, Karnataka, India

• Reduced feature document retrieval time by 80% for 3-4 internal teams by designing an infrastructure to filter and serve Markdown documents as structured HTML with consistent CSS styling.

Harsh’s AI Stack

claude code
claude code
codex
codex
cursor
cursor
deepseek
deepseek
gemini
gemini
github copilot
github copilot

Education

Dayananda Sagar College of Engineering, BANGALORE

Bachelor of Engineering, Computer and Information Sciences and Support Services

Dayananda Sagar College of Engineering, BANGALORE

20154 yrs2019

Join Harsh and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Harsh Kumar on Tal

Headline

Senior Software Engineer at Tellius | Building RAG & LLM-Orchestration Systems (LangGraph, Qdrant, Multi-Agent Retrieval)

About

I'm a Senior Software Engineer at Tellius, where I design and build the AI infrastructure behind our decision-intelligence platform — from production-grade Retrieval-Augmented Generation (RAG) systems to autonomous quality-monitoring agents. Most recently, I architected and independently built tellius-health-agent, an autonomous service that auto-discovers a customer deployment's data and configuration and runs automated + LLM-judge evaluations across five capability areas (Text-to-SQL, RAG, routing, MCP, regression) — cutting manual QA effort from hours to minutes and shrinking failure-detection time from customer-reported (days) to a nightly cycle. Before that, I led a cross-timezone team of 3 engineers to ship a production RAG platform spanning 5 enterprise document sources (Google Drive, S3, GCS, Azure Blob, SharePoint), combining custom Airbyte connectors, Mistral OCR, Airflow-orchestrated hybrid Qdrant indexing (BM25 + SPLADE + Dense), and a LangGraph multi-agent retrieval service with RRF fusion and cross-encoder reranking. Earlier in my career at Tellius and at Lam Research, I built backend APIs, ETL pipelines, and production tooling — including infrastructure that cut document retrieval time by 80% and deadlock resolution time by 90%. I work primarily in Python, with FastAPI, LangGraph, LangChain, Qdrant, Airflow, and Kubernetes — always happy to talk applied LLM systems, RAG architecture, or agentic evaluation.

Location

Greater Bengaluru Area

AI stack

claude code

codex

cursor

deepseek

gemini

github copilot

Work experience

Senior Software Engineer at Tellius, Aug 2025 - Present · 1 yr 2 mos, Bengaluru · Hybrid, • Architected and independently built tellius-health-agent, an autonomous FastAPI service that auto-discovers a deployment's Business Views, unstructured workspaces, and workflows, and runs automated + LLM-judge evaluations across 5 capability areas (Text-to-SQL, RAG, routing, MCP, regression) — cutting per-deployment QA time from hours to minutes and shrinking failure-detection latency from customer-reported (days) to a nightly automated cycle. • Authored the technical design and phased implementation roadmap for company-wide AI-quality monitoring spanning structured, unstructured, routing, and MCP testing areas. • Architected and shipped a production RAG platform across 5 enterprise document sources (Google Drive, S3, GCS, Azure Blob, SharePoint) by leading a cross-timezone team of 3 engineers to deliver custom Airbyte connectors with Mistral OCR, Airflow-orchestrated hybrid Qdrant indexing (BM25 + SPLADE + Dense), and a LangGraph multi-agent retrieval service with RRF fusion and cross-encoder reranking.

Software Engineer 2 at Tellius, Feb 2022 - Jul 2025 · 3 yrs 6 mos, Bengaluru · Hybrid, • Reduced query processing time for moving window calculations by 30% by redesigning the aggregation architecture to support moving average, moving min, and moving sum. • Enabled multilevel sorting and top-N filtering for pivot tables, reducing report generation stepsby 80%, by architecting an extensible pivot table engine. • Improved operational visibility for 5-6 background jobs by designing a real-time task tracking system, reducing mean debugging time by 50%.

Software Engineer ll at Lam Research, Feb 2021 - Feb 2022 · 1 yr 1 mo, Bangalore Urban, • Reduced mean time to resolve deadlocks by 90% by implementing automated deadlock detection that captures detailed logs and tool state on trigger. • Improved team technical proficiency across 3 technology domains by leading regular knowledge-sharing sessions, driving continuous learning within the team. • Improved tool throughput by resolving critical escalations and performance bottlenecks identified through systematic profiling.

Software Engineer at Lam Research, Aug 2019 - Feb 2021 · 1 yr 7 mos, Bengaluru, Karnataka, India, • Reduced feature document retrieval time by 80% for 3-4 internal teams by designing an infrastructure to filter and serve Markdown documents as structured HTML with consistent CSS styling.

Education

Bachelor of Engineering, Computer and Information Sciences and Support Services, Dayananda Sagar College of Engineering, BANGALORE