Debanshu Mutt profile photo 1 on Tal

Debanshu Mutt

SDE-2 | Backend Engineer | Python | LLMs | Agentic AI | AI Orchestration | Distributed Systems | Ex-H&M / Maersk

Bengaluru

5+ years of work experience

Publicis.Sapient
Jun 20264 moPresent

Bengaluru

Contributed to the AI-powered AppMod platform using Python, FastAPI, Agno agents, LLMs, and LanceDB, engineering code modernization workflows that reduced costs by up to 50%. • Enhanced the AI chat module with intent classification, tool scoping, and lexical ranking to limit retrieval to the top 30 capabilities, reducing latency by 30% and LLM retry failures by 65%. • Integrated Redis-backed WebSocket connection management with polling fallback, reducing redundant frontend polling by 80% and enabling real-time updates for large-repository async workflows. • Engineered cooperative cancellation across code ingestion, spec, and Epic/Story workflows, reducing Cancel API latency to under 500ms and avoiding 40% of downstream LLM calls while preserving partial artifacts. • Built per-step LLM token telemetry for the code ingestion pipeline -tracking input, output, cached, and failed tokens- identifying optimizations that reduced LLM token consumption by 15%.
H&M
Jul 20241 y 9 moApr 2026

Bangalore Urban

Built a scalable event-driven system with FastAPI handling 3M+ requests daily, syncing PLM data in real time via RabbitMQ, Spanner, Kafka and Solace. • Refactored over 3,000 lines of legacy code using SOLID principles, reducing bug resolution time by 75%. • Designed and owned a Python-based data ingestion pipeline (Postgres - BigQuery) deployed on Kubernetes with dynamic configs, improving data availability for 5+ teams and reducing manual effort by 60%. • Built a GCP-powered logging pipeline from Cloud Run to Log Sink and Pub/Sub, powering Splunk dashboards; improved monitoring coverage by 55% and reduced incident response time by 40%. • Built a secure LangGraph-powered FastMCP backend that orchestrated intent classification, tool selection, and SQL generation for PLM analytics, enforcing OAuth-based supplier isolation across 5+ teams and reducing ad-hoc requests by 70%. • Developed shared Python utilities and mentored juniors via PR reviews, reducing duplication and speeding up feature delivery by 25%.
A.P. Moller - Maersk
May 20222 yMay 2024

Bengaluru

Developed a scalable backend service for the GenAI Agent Assist project using Python, FastAPI, LangChain, and Azure Functions. Orchestrated intent classification, RAG, and MCP tool workflows to generate LLM- powered recommendations, processing 50K daily requests and saving $4M annually. • Engineered a robust REST API solution for the Pricing Team using FastAPI and MySQL. Significantly reduced data retrieval time by 70% through optimized database queries and the implementation of data partitioning logic. • Designed a Flask API using OCR and OpenCV to automate release order document authorization, leveraging multiprocessing and PyMuPDF to reduce response time from 40 to 5 seconds. • Implemented CI/CD pipelines using GitHub Actions, streamlining development and deployment. Deployed APIs in Docker containers, leveraging containerization for efficient scaling and maintenance. • Utilized Microsoft Azure Cloud (Web Apps, Function Apps, ACR) services and rule-based scalability to host and manage APIs, ensuring efficient handling of fluctuating loads while optimizing costs.
A.P. Moller - Maersk
Jun 20211 yrMay 2022

• Automated diverse business use cases utilizing Java, Apache POI, Sikuli, and Selenium, leading to annual savings of around $5M for the company. • Utilized PowerBI and DAX measures to create visuals, providing pricing managers insights into week-to-week volume pressure and pricing patterns. Maintained data pipelines using SQL, Databricks, etc., resulting in a 30% reduction in manual effort.

Debanshu’s AI Stack

claude code
claude code
cursor
cursor
hugging face
hugging face
langchain
langchain
langgraph
langgraph
mcp
mcp

Education

KIIT - Kalinga Institute of Industrial Technology

Bachelor of Technology - BTech, Computer Science and Communications

KIIT - Kalinga Institute of Industrial Technology

20184 yrs2022

High School Diploma

DBMS English School

200414 yrs2018

Join Debanshu and 10k+ other professionals on

Tal Boss

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

Download Tal Boss

Debanshu Mutt on Tal

Headline

SDE-2 | Backend Engineer | Python | LLMs | Agentic AI | AI Orchestration | Distributed Systems | Ex-H&M / Maersk

About

Backend Engineer with 4.5+ years of experience building scalable, cloud-native distributed systems across retail and logistics. I specialize in Python/ FastAPI/ Flask, designing event-driven microservices and data platforms that handle millions of requests daily. Recently, I’ve worked extensively on production-grade Generative AI systems, building LLM-powered workflows using LangChain, FAISS, Redis, and Azure/GCP. My experience includes RAG pipelines, secure MCP-based data access, and scalable AI-driven APIs handling 50k+ daily requests. Strong in distributed systems, Kubernetes, messaging platforms (Kafka, RabbitMQ), I focus on building resilient, observable, and high-impact backend and AI platforms.

Location

Bengaluru

AI stack

claude code

cursor

hugging face

langchain

langgraph

mcp

Work experience

Software Engineer at Publicis.Sapient, Jun 2026 - Present · 4 mo, Bengaluru, Contributed to the AI-powered AppMod platform using Python, FastAPI, Agno agents, LLMs, and LanceDB, engineering code modernization workflows that reduced costs by up to 50%. • Enhanced the AI chat module with intent classification, tool scoping, and lexical ranking to limit retrieval to the top 30 capabilities, reducing latency by 30% and LLM retry failures by 65%. • Integrated Redis-backed WebSocket connection management with polling fallback, reducing redundant frontend polling by 80% and enabling real-time updates for large-repository async workflows. • Engineered cooperative cancellation across code ingestion, spec, and Epic/Story workflows, reducing Cancel API latency to under 500ms and avoiding 40% of downstream LLM calls while preserving partial artifacts. • Built per-step LLM token telemetry for the code ingestion pipeline -tracking input, output, cached, and failed tokens- identifying optimizations that reduced LLM token consumption by 15%.

Software Engineer (SDE-2) at H&M, Jul 2024 - Apr 2026 · 1 y 9 mo, Bangalore Urban, Built a scalable event-driven system with FastAPI handling 3M+ requests daily, syncing PLM data in real time via RabbitMQ, Spanner, Kafka and Solace. • Refactored over 3,000 lines of legacy code using SOLID principles, reducing bug resolution time by 75%. • Designed and owned a Python-based data ingestion pipeline (Postgres - BigQuery) deployed on Kubernetes with dynamic configs, improving data availability for 5+ teams and reducing manual effort by 60%. • Built a GCP-powered logging pipeline from Cloud Run to Log Sink and Pub/Sub, powering Splunk dashboards; improved monitoring coverage by 55% and reduced incident response time by 40%. • Built a secure LangGraph-powered FastMCP backend that orchestrated intent classification, tool selection, and SQL generation for PLM analytics, enforcing OAuth-based supplier isolation across 5+ teams and reducing ad-hoc requests by 70%. • Developed shared Python utilities and mentored juniors via PR reviews, reducing duplication and speeding up feature delivery by 25%.

Associate Full Stack Engineer at A.P. Moller - Maersk, May 2022 - May 2024 · 2 y, Bengaluru, Developed a scalable backend service for the GenAI Agent Assist project using Python, FastAPI, LangChain, and Azure Functions. Orchestrated intent classification, RAG, and MCP tool workflows to generate LLM- powered recommendations, processing 50K daily requests and saving $4M annually. • Engineered a robust REST API solution for the Pricing Team using FastAPI and MySQL. Significantly reduced data retrieval time by 70% through optimized database queries and the implementation of data partitioning logic. • Designed a Flask API using OCR and OpenCV to automate release order document authorization, leveraging multiprocessing and PyMuPDF to reduce response time from 40 to 5 seconds. • Implemented CI/CD pipelines using GitHub Actions, streamlining development and deployment. Deployed APIs in Docker containers, leveraging containerization for efficient scaling and maintenance. • Utilized Microsoft Azure Cloud (Web Apps, Function Apps, ACR) services and rule-based scalability to host and manage APIs, ensuring efficient handling of fluctuating loads while optimizing costs.

Full Stack Engineer Intern at A.P. Moller - Maersk, Jun 2021 - May 2022 · 1 yr, • Automated diverse business use cases utilizing Java, Apache POI, Sikuli, and Selenium, leading to annual savings of around $5M for the company. • Utilized PowerBI and DAX measures to create visuals, providing pricing managers insights into week-to-week volume pressure and pricing patterns. Maintained data pipelines using SQL, Databricks, etc., resulting in a 30% reduction in manual effort.

Education

Bachelor of Technology - BTech, Computer Science and Communications, KIIT - Kalinga Institute of Industrial Technology

High School Diploma, DBMS English School