Sudhanshu B.

AI Researcher

Bengaluru, Karnataka, India4 yrs 8 mos experience

Most Likely To SwitchAI ML Practitioner

Key Highlights

Expertise in Generative AI and document intelligence.
Published research at top-tier conferences.
Strong foundation in machine learning and software development.

Stackforce AI infers this person is a SaaS-focused AI/ML engineer with strong software development skills.

Contact

Skills

Core Skills

Generative AiSoftware DevelopmentSystems DesignResearch Skills

Other Skills

Large Language Models (LLM)Python (Programming Language)C++Deep LearningMachine LearningiOS DevelopmentJavaSwift (Programming Language)Computer EngineeringPitching IdeasAndroid DevelopmentNatural Language Processing (NLP)Computer VisionComputer ScienceTensorFlow

About

I am a computer engineering student who loves to think through problems and build creative solutions. Enthusiast in field of AI, currently, I am studying machine learning and applying models to create a better world.

Experience

4 yrs 8 mos

Total Experience

2 yrs 3 mos

Average Tenure

4 yrs 3 mos

Current Experience

Amazon

3 roles

Applied Scientist 2

Promoted

Oct 2025 – Present · 6 mos

Systems DesignGenerative AILarge Language Models (LLM)Research SkillsSoftware Development

Software Development Engineer 2

Dec 2023 – Oct 2025 · 1 yr 10 mos

Architected 'DigiScan', a highly configurable document extraction engine that achieved >90% accuracy. Designed the system with a layered architecture to ensure extensibility, enabling the rapid onboarding of diverse Amazon business lines with minimal configuration overhead.
Led the org-wide scientific strategy for productionizing Generative AI, developing methodologies to solve for explainability, latency, and cost. Pioneered Visual Grounding techniques and Human-in-the-Loop workflows to ensure high-precision extraction. Trained specialized Small Language Models (SLMs) that matched the accuracy of larger foundation models, resulting in a significant reduction in inference costs and processing latency.
Enhanced the document suite’s capabilities by engineering solutions for non-textual entity extraction, including custom QR code models and advanced document classification algorithms to ensure holistic data capture.
Established thought leadership in the domain by publishing 2 papers at A* rated conferences (KDD, AAAI) and 4 at internal conference (AMLC), validating novel approaches in document intelligence.

Software Development Engineer 1

Jan 2022 – Dec 2023 · 1 yr 11 mos

Designed and implemented vendor invoice management solution from scratch to perform document ingestion, compliance, extraction of metadata, and scanning for virus (DICES)
Designed and implemented document extraction service (DigiScan) using OCR

Genesys

2 roles

Associate AI/ML Engineer

Aug 2021 – Dec 2021 · 4 mos

Intent Miner is a service used to get intents from transcripts which are then used to develop Bot workflows. Made backend changes, including tests, to ease addition of multiple languages and dialects. Deployed few languages and dialects in production.
Refactored code to improve modularity and wrote tests for various Topic Miner modules.
Worked on NER service to investigate special cases in the ML model prediction.

Artificial Intelligence Intern

Jun 2021 – Jul 2021 · 1 mo

Researched, developed and deployed a Smart Reply system posed as an Information Retrieval problem using parallel T5 transformers, Symmetric Loss and other language models like GPT-3, BERT for intent, sentiment and LM error to get top-n replies for a given message.

Upcloud technology

DL Intern

Jul 2020 – Sep 2020 · 2 mos · Mumbai, Maharashtra, India

Designed and developed a ML system for a period tracker application. Consisted of an online ML model trained to adapt predictions to a user and a global model to improve base/initial model predictions.

Erfinden

Technology Intern

May 2020 – Jul 2020 · 2 mos · Pune, Maharashtra, India

Magento is an open-source e-commerce platform for a vendor to sell products online. Designed and developed a system around Magento, using AMP stack, for multiple vendors to sell products, and receive organised customer orders.

Nvidia

Student Intern

Dec 2019 – Mar 2020 · 3 mos · Pune Area, India

FFmpeg: CUDA implementation for hstack and vstack filters in audio/video filtering library - libavfilter. Involved understanding structure and working of ffmpeg filters, parallel computing concepts, and CUDA (advanced C) programming.

National institute of opthalmology

Research Intern

May 2019 – Aug 2019 · 3 mos · Pune Area, India

Detecting FAZ Area from OCTA scan of human retina using U-Net.
Segmenting FAZ area and studying its changes over time can help ophthalmologists detect various defects in human retina. Used U-Net, a CNN developed for biomedical image segmentation, for segmenting FAZ area from OCTA scans (Dataset: OCTAGON).
Worked on finding and justifying capillary density value in an OCTA scan.