Amrit Sinha

AI Researcher

Bengaluru, Karnataka, India0 mo experience

AI EnabledAI ML Practitioner

Key Highlights

Passionate about fintech and technology integration.
Experienced in AI and data analysis for business optimization.
Strong problem-solving skills demonstrated in real-world projects.

Stackforce AI infers this person is a Fintech-focused AI developer with strong analytical skills.

Contact

Skills

Core Skills

Natural Language Processing (nlp)Deep Learning

Other Skills

AI modelsArtificial Intelligence (AI)Artificial Intelligence for BusinessC++CUDAComputer ScienceConvolutional Neural Networks (CNN)Data AnalysisElectronic TradingEquity TradingFinancial MarketsGenerative AIMarket DataProblem SolvingProduct Development

About

codechef link: https://www.codechef.com/users/omlettesinha Hello! I'm a Computer Science and Engineering student at RNSIT, passionate about technology, finance, and economics. I want to build and use my skills to break-in fintech domain. I have a keen interest in data, algorithms, market trends and economic decision-making. Outside academics, I'm an avid reader dedicated to staying current in these rapidly evolving industries. I have a 'Debater' personality trait, which fuels engaging discussions. I'm eager to connect with like-minded individuals, professionals, and enthusiasts in the fields of technology, finance, economics, and psychology. Let's network, collaborate, and explore these exciting domains together!"

Experience

0 mo

Total Experience

Average Tenure

Current Experience

Sumeru global support solutions pvt. ltd.

AI Intern

Sep 2025 – Present · 8 mos · Bengaluru, Karnataka, India

Architected a dual-GPU AI pipeline using Whisper and Pyannote to transcribe Hinglish calls, visualizing Agent-vs-Customer engagement trends via smoothed "Interest Graphs" for sales optimization.
Developed a Python pipeline to analyze customer call recordings using multiple AI models.
Used openai/whisper-large-v3 for speech-to-text transcription of "Hinglish" audio.
Integrated Hubert (Emotion) and two BERT/XLM-R (Sentiment) models to analyze audio tone and text content.
Solved critical CUDA out of memory errors by architecting a multi-script, "Divide and Conquer" workflow.
Created a dedicated preprocessing script to clean and denoise all audio files using librosa and noisereduce.
Combined emotion and sentiment scores into a final 0-10 call quality rating, visualized with matplotlib.

Natural Language Processing (NLP)Deep LearningPython (Programming Language)Data AnalysisConvolutional Neural Networks (CNN)AI models+4