Animesh Kumar
Contact
01Overview

Eight years in, the work has narrowed to one problem: making software that can prove what it says. At Adobe I build attribution1 for Acrobat's AI Assistant2 — the machinery that traces every sentence a model writes back to the exact span of the document it came from. Before that, Azure infrastructure3 for Microsoft's reporting stack. On weekends, GoWeekender4.

Read the CV
02Filings3 pending
03Record2018 — present

Adobe

Noida, India · March 2021 — present

I work on the trust layer of Adobe's document AI: the systems that decide what an assistant is allowed to claim, and prove where each claim came from. That has meant shipping AI Assistant to Acrobat across languages, filing three patents on attribution and summarization, and cutting the cost of running these models in production.

  • Python
  • PyTorch
  • LLM inference
  • AWS
  • React
  • Node.js
Full record→
  1. Jan 2025 — present
    Computer Scientist II (SDE3)Lead attribution work for document question answering · Filed two patents in the last eighteen months
  2. Aug 2022 — Dec 2024
    Computer Scientist I (SDE2)Built AI Assistant features across multiple languages · Scaled AI Assistant to 10M+ users
  3. Mar 2021 — Jul 2022
    Member of Technical Staff IIBuilt an ML model optimization framework · Cut inference time 43.5% and cost 52%

MAQ Software (Microsoft)

Hyderabad, India · Sept 2018 — Feb 2021

Migrated an on-premise database to Azure and designed the cloud architecture around it — Azure SQL DB, Databricks, elastic pools, elastic search — tuning SQL DB until the reports built on top of it were fast enough to use. Built the web application layer with .NET, MVC, AngularJS, TypeScript and the Power BI API, backed by Redis, file storage and KeyVault.

  • C#
  • .NET
  • Azure SQL
  • Databricks
  • Power BI
  • TypeScript
Full record→
  1. Mar 2020 — Feb 2021
    Software Engineer IILed a team of five engineers · Optimized cloud infrastructure and cost
  2. Sept 2018 — Feb 2020
    Software Engineer IImplemented a distributed logging system · Reduced deployment time by 50%
04Work3 entries
Adobe2022 — present

Acrobat AI Assistant

Conversational reading for PDFs, with every answer traced back to the page it came from.

10M+ users; multilingual rollout
Adobe2021 — 2022

Model optimization framework

A framework for making production ML models cheap enough to leave switched on.

−43.5% inference time, −52% cost
MAQ Software2018 — 2020

Azure platform migration

Moved a reporting stack off on-premise SQL and onto Azure without losing a night's sleep.

−35% cloud spend, 99.99% reliability
05WeekendsGoWeekender

Travel is what I organise the rest of my year around.

Animesh Kumar
Fig. 1 — Somewhere worth the weekend

GoWeekender is what I do with that on weekends. It matches you with ID-verified travellers who share your vibe, your budget and your pace, so a weekend trip starts with the right people instead of a group chat that never leaves the planning stage. Go solo, plan something yourself, or join an escape someone else is hosting.

It came out of years of being planned for by strangers — the write-ups, routes and warnings that got me out of the city and back again. This is the part where I give some of that back. It isn't a business and it isn't trying to become one: I build it in the evenings and on Sundays, which means it moves slowly, and I'd rather it moved slowly than became something else.

  • Vibe Check — matched on adventure style, budget and pace
  • Government-ID verification for every member
  • Women-only tribes, live SOS, Ghost Mode
  • Go solo, plan a group yourself, or join a host-led escape
  • Chat, itineraries and split payments built in

The only promise

The app is free, forever. No ads, no subscription, no paywall three screens in. Trips cost what trips cost; the app never will.

goweekender.in↗
06Marginalia

Recognition

  • Adobe Impact AwardTechnical excellence, Document Cloud2023
  • Technical Excellence AwardsFour, across Adobe and MAQ Software2018 — present
  • NASSCOM Smart City HackathonRank 3, NMIT region2017
  • National Cyber OlympiadState rank 4 · national rank 5,5142015
  • National Mathematics OlympiadState rank 11 · national rank 3642015

Education

SJB Institute of Technology, Bangalore

B.Tech, Computer Science and Engineering

2014 — 2018

Graduated first class, specializing in machine learning and backend development.

  • Started the campus Linux enthusiasts' club
  • Rank 3, NASSCOM Smart City Hackathon (NMIT region)
  • Led the college quiz team through inter-college competition
sjbit.edu.in ↗