Garmit Pant profile photo 1 on Tal

Garmit Pant

SDE-1 at Amazon | Ex-MathWorks | IIIT Delhi

Bengaluru

3+ years of work experience

Amazon
May 20251 yr 5 mosPresent

Bengaluru

• Architected and delivered end-to-end agentic subscription management for Alexa for Shopping, Amazon's AI assistant (129M+ MAU), replacing static help flows with live subscription data across 8.5M+ annual queries and 25+ digital businesses. • Designed and shipped MCP tools integrated with Amazon's centralized multi-tenant subscription platform, federating 30+ backend services to surface real-time billing, renewal, and plan data conversationally. • Built layered intent disambiguation for agentic tool routing, resolving cross-domain conflicts to hold resolution to under 2 tool invocations per query and contain per-query latency and inference cost. • Drove end-to-end productionization of 20+ agentic user actions from POC through phased rollout, and built an LLM-as-judge evaluation harness with configurable judge prompts to surface agent quality regressions before launch. • Owned and led migration of the Prime Video subscription channel data service (487M monthly requests, 1.6B at peak) off a legacy dependency, building an asynchronous validation framework that verified response parity at zero customer latency impact. • Drove an overhaul of the unit-testing framework across the entire hydrator test suite, refactoring dummy stubs into mock clients of downstream services so tests exercised real call paths rather than placeholder responses. • Executed load testing across 6 peak-simulation GameDays (NA/EU/FE) and drove capacity-scaling analysis behind zero-outage Prime Day 2025 traffic (11.2M customer visits), including scaling a distributed ElasticSearch cluster from 4 to 20 data nodes. • Architected the orchestration engine and agent scaffold for an internal multi-agent AI platform (now 22 agents and 29 skills across 2 teams) that automates availability root-cause analysis and weekly on-call reporting, saving roughly 740 engineer-hours per year.
MathWorks
Jul 20231 yr 2 mosAug 2024

Hyderabad

• Developed and validated an end-to-end object detection and tracking pipeline for video built on vision foundation models (Grounding-DINO, SAM, DeAOT), with significant Python development — improving tracking accuracy by 8% and reducing user touchpoints by 50%. • Improved semantic segmentation accuracy on hyperspectral imagery from 74% to 96% using U-Net, through extensive data pre-processing and post-processing, augmentation, and hyperparameter optimization across MATLAB and Python. • Built a parallelized multiprocessing data pipeline in Python that accelerated Radiative Transfer Modelling-based atmospheric correction of ~650GB of satellite imagery by 4x — designing the workflow end to end, and driving significant C++ code review and debugging. • Engineered an internal operations dashboard (React, Node.js, Express, MongoDB) with RESTful APIs and role-based data models, reducing onboarding time by 30% for 500+ employees. • Contributed to the MATLAB community by solving 100+ queries on the MATLAB Answers forum.
MathWorks
May 20223 mosJul 2022

Hyderabad, Telangana, India

MIDAS: Multimodal Digital Media Analysis Lab
Jan 202110 mosOct 2021

Garmit’s AI Stack

claude code
claude code
claude
claude
kiro
kiro
langchain
langchain
mcp
mcp
OpenRouter
OpenRouter

Education

Indraprastha Institute of Information Technology, Delhi

Bachelor of Technology - BTech, Computer Science and Applied Mathematics

Indraprastha Institute of Information Technology, Delhi

20194 yrs2023
Delhi Public School Rohini

High School Diploma, Science (Non-medical)

Delhi Public School Rohini

20172 yrs2019

Join Garmit and 10k+ other professionals on

Tal Boss

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

Download Tal Boss

Garmit Pant on Tal

Headline

SDE-1 at Amazon | Ex-MathWorks | IIIT Delhi

About

Software engineer at Amazon building agentic AI systems that run in production at scale — currently on Alexa for Shopping, Amazon's AI assistant serving 129M+ monthly users. My work sits where LLM systems meet distributed backend engineering: multi-agent orchestration, intent and tool routing, evaluation harnesses that keep agent quality honest, and the services underneath handling hundreds of millions of requests a month. Before Amazon I worked in computer vision at MathWorks — semantic segmentation, object detection and tracking with vision foundation models — which is where I learned what it takes to move a model from a paper to something that holds up on real data. I'm highly inquisitive and do my best work when exploration runs straight into hands-on implementation. Always happy to talk about agents, evals, or anything at the boundary of AI and systems.

Location

Bengaluru

AI stack

claude code

claude

kiro

langchain

mcp

OpenRouter

Work experience

Software Development Engineer I at Amazon, May 2025 - Present · 1 yr 5 mos, Bengaluru, • Architected and delivered end-to-end agentic subscription management for Alexa for Shopping, Amazon's AI assistant (129M+ MAU), replacing static help flows with live subscription data across 8.5M+ annual queries and 25+ digital businesses. • Designed and shipped MCP tools integrated with Amazon's centralized multi-tenant subscription platform, federating 30+ backend services to surface real-time billing, renewal, and plan data conversationally. • Built layered intent disambiguation for agentic tool routing, resolving cross-domain conflicts to hold resolution to under 2 tool invocations per query and contain per-query latency and inference cost. • Drove end-to-end productionization of 20+ agentic user actions from POC through phased rollout, and built an LLM-as-judge evaluation harness with configurable judge prompts to surface agent quality regressions before launch. • Owned and led migration of the Prime Video subscription channel data service (487M monthly requests, 1.6B at peak) off a legacy dependency, building an asynchronous validation framework that verified response parity at zero customer latency impact. • Drove an overhaul of the unit-testing framework across the entire hydrator test suite, refactoring dummy stubs into mock clients of downstream services so tests exercised real call paths rather than placeholder responses. • Executed load testing across 6 peak-simulation GameDays (NA/EU/FE) and drove capacity-scaling analysis behind zero-outage Prime Day 2025 traffic (11.2M customer visits), including scaling a distributed ElasticSearch cluster from 4 to 20 data nodes. • Architected the orchestration engine and agent scaffold for an internal multi-agent AI platform (now 22 agents and 29 skills across 2 teams) that automates availability root-cause analysis and weekly on-call reporting, saving roughly 740 engineer-hours per year.

Software Engineer at MathWorks, Jul 2023 - Aug 2024 · 1 yr 2 mos, Hyderabad, • Developed and validated an end-to-end object detection and tracking pipeline for video built on vision foundation models (Grounding-DINO, SAM, DeAOT), with significant Python development — improving tracking accuracy by 8% and reducing user touchpoints by 50%. • Improved semantic segmentation accuracy on hyperspectral imagery from 74% to 96% using U-Net, through extensive data pre-processing and post-processing, augmentation, and hyperparameter optimization across MATLAB and Python. • Built a parallelized multiprocessing data pipeline in Python that accelerated Radiative Transfer Modelling-based atmospheric correction of ~650GB of satellite imagery by 4x — designing the workflow end to end, and driving significant C++ code review and debugging. • Engineered an internal operations dashboard (React, Node.js, Express, MongoDB) with RESTful APIs and role-based data models, reducing onboarding time by 30% for 500+ employees. • Contributed to the MATLAB community by solving 100+ queries on the MATLAB Answers forum.

EDG Intern at MathWorks, May 2022 - Jul 2022 · 3 mos, Hyderabad, Telangana, India

Research Intern at MIDAS: Multimodal Digital Media Analysis Lab, Jan 2021 - Oct 2021 · 10 mos

Education

Bachelor of Technology - BTech, Computer Science and Applied Mathematics, Indraprastha Institute of Information Technology, Delhi

High School Diploma, Science (Non-medical), Delhi Public School Rohini

Garmit Pant - Software Development Engineer I in Bengaluru | Tal