Siddharth Kumar

Software Engineer

Bengaluru, Karnataka, India8 yrs 11 mos experience

Most Likely To SwitchAI Enabled

Key Highlights

Expert in machine learning and computer vision technologies.
Proven track record in developing scalable software solutions.
Strong experience in localization and translation services.

Stackforce AI infers this person is a SaaS expert with a strong focus on machine learning and computer vision.

Contact

Skills

Core Skills

Software InfrastructureLarge Language Models (llm)Machine LearningComputer Vision

Other Skills

Time Series AnalysisTensorFlowspannerGoogle GeminiApache KafkaiOSJavaCAlgorithmsC++EnglishPython (Programming Language)Big DataData Engineering

About

Working at Google, GStore team, on google-wide localization service. Past: Google, Workspace search-quality team, on gmail and chat search. Goldman Sachs, building data-processing pipelines to automate production environment monitoring. Adobe, in Adobe-Scan as an ML engineer, shipping deep learning models in computer vision domain.

Experience

8 yrs 11 mos

Total Experience

2 yrs 11 mos

Average Tenure

4 yrs 10 mos

Current Experience

Google

Software Engineer 3

Aug 2021 – Present · 4 yrs 10 mos · Bengaluru, Karnataka, India · On-site

Working with GStore Localization team.
Some projects I worked on:
1. Notification Service for Translation Management.
2. AutoML Machine Translations.
3. Gemini-based Language Quality Analytics Dashboard.
4. Client-specific Translation Memory System.
5. Multi-provider Translation API.
Previously I worked with Google Workspace Search quality to make emails and chats more findable for the users.
Some of the projects I'm proud of:
1. ”Did You Mean” Spell Suggestions in Google Chat.
2. Relevance Ranking for Google Chat search.
3. Several projects in enhancing query rewrite capabilities for gmail search.

Software InfrastructureLarge Language Models (LLM)

Goldman sachs

Associate

Aug 2020 – Aug 2021 · 1 yr · Bengaluru, Karnataka, India · Remote

Building machine learning pipelines to automate the monitoring of the production environment.

Machine LearningTime Series Analysis

Adobe

3 roles

Senior Member Of Technical Staff (SDE-2)

Promoted

Jan 2019 – Aug 2020 · 1 yr 7 mos · Noida, India

ML Research Collaboration:
Led tech transfer of an edge detection model from Adobe Research, achieving 10% accuracy gain.
Document Detection:
Applied transfer-learning on MobileNet, delivering one of the app’s top-3 most loved features.
Text Detection:
Developed an ML model to enhance text-darkening and image quality for scanned documents.
ML-On-Device SDK:
Integrated Adobe’s ML SDK for Android, reducing inference time by 20% and enabling cross-team adoption.

Computer VisionTensorFlow

Member Of Technical Staff (SDE-1)

Jun 2017 – Dec 2018 · 1 yr 6 mos · Noida, India

Document Edge Detection:
Developed a real-time document edge detection deep learning model for Adobe Scan which works on image-to-image translation.
Using deep learning fused with a heuristic approach (Patent ID - us P8681-US/AD01.P8681), lead to ~15% increase in detected corners accuracy
Some ideas I worked on include –
1. Optimization of convolution layers, replacing some by depth-wise separable convolutions.
2. Signed-distance-transformation on dataset, to get the segmentation of documents.
3. Two decoder heads on a single encoder, one for document edge, and other for document segmentation.
4. Created synthetic dataset for documents, to have a reasonable size and representation of training data.
5. Fine-tuning encoder-decoder architecture to make up for the lack of real-world dataset.
6. Experimented with different loss functions – and came up with effective losses for document segmentation purposes, joint effort with Adobe research team.

Computer VisionTensorFlow

Product Intern

May 2016 – Jul 2016 · 2 mos · Noida, India

Implemented a framework to use machine learning for partitioning a PDF page into logical blocks (paragraphs, lists and bordered tables).
Used genetic algorithm for improving the logical bounds obtained.