Suryansh Singh Rawat
AI Engineer in Bangalore, India
About
Suryansh Singh Rawat holds a Bachelor of Engineering in Electronics and Instrumentation from BITS Pilani. Currently, he is an AI Engineer-II at Ascentt, where he has contributed to developing scalable Generative AI solutions for enterprises. Previously, Suryansh worked as a Data Scientist at Deloitte USI, where he built AI solutions for cyber threat detection using custom Autoencoder architectures and NLP-based generative models. During his internship at Scienaptic AI, Suryansh gained experience in developing credit underwriting models using XGBoost and Logistic Regression, improving application approval rates.
Suryansh has worked on various projects, including building a multilingual RAG QA model and developed a deep learning-based system to detect GAN-generated deepfake images. He also gained research experience at IIASA, Vienna, where he contributed to a Plant-FATE simulation model by integrating C++ code for plant simulations under varying climate conditions.
With technical skills in Python, SQL, C++, and experience with frameworks like TensorFlow, PyTorch, and LangChain, Suryansh continues to apply his expertise in AI and machine learning to build impactful solutions across various domains.
Education
Electronics and Instrumentation Engineering
Work Experience
-
Spearheaded the development of ToyotaGPT, an enterprise chat platform, designed REST APIs in Python, implemented user data management with DynamoDB, and built a RAG system for analyzing diverse file types, scaling the platform to serve 35,000+ active users.
-
Assembled a metadata-extracting chatbot using LangChain and Pydantic to enable precise car model searches in PGVector DB. Implemented a Dynamic RAG Agentic approach to route vector DB collections based on query type and complexity.
-
Designed an Agent using a ReAct framework for web browsing using DuckDuckGo API, dynamically routing user queries between vector DBs and internet searches based on query intent.
-
Created DocuBot, an end-to-end chatbot for analyzing tabular contracts using a custom Pandas DataFrame Agent.
-
Leveraged Deep Learning NLP techniques to develop Deloitte IRL tool (multi-document to single-document requirements library automation tool). Created multiclass ensemble classification models and unsupervised clustering models with HDBSCAN and t-SNE, finetuned document embeddings with Metric Learning using Sub-center ArcFace Loss.
-
Constructed a Text-Summarization Model for generating concise cyber threat reports using SecBERT. Incorporated NER Tagging and Masking models with SpaCy annotation. Developed model pipeline leveraging Flask and Docker for efficient deployment.
-
Developed custom Autoencoder architectures along with Graph Network based preprocessing with Neo4J to identify anomalies for detecting Zero-Day Cyber Threats using realtime cloud network flow data.
-
Built Generative AI solutions for a regulatory compliance tool by creating RAG powered Entity Extraction and Summarization models and exploited LLM models like Llama 2 and GPT-4, orchestrated pipelines using Langchain and deployed models using Streamlit.
-
Leveraged XGBoost and Logistic Regression to create Credit Underwriting ML Models to forecast the probability of credit accounts that may default, enabling an 11.5% increase in application approvals.
-
Designed a Unified Credit Underwriting model framework leveraging 100MM+ raw credit records, deployed to deliver baseline performance metrics for credit unions across diverse risk segments.
-
Developed Customer-Employee classification model for security surveillance using YOLOv4 and TensorFlow.
-
Created a unified pipeline to transcribe webinar audio into text and derived insights by extracting topics using LDA topic modeling.
-
Built medical text summarization model which used BioBERT along with RAKE Algorithm to generate key phrases.
Projects
-
Developed a QA RAG model using OpenAI and Langchain. Utilized Qdrant vector DB and Cohere embeddings for multilingual capabilities.
-
Successfully deployed and hosted the model with a Streamlit UI, integrated a speech-to-speech feature for user interaction.
- Developed a Convolutional Neural Network (CNN) architecture to efficiently detect GAN-generated DeepFake images.
- Achieved values of precision and recall as 0.9843 and 0.9710 respectively with an accuracy score of 97.77% using StyleGAN dataset.
- Developed a CNN model for music genre classification using libraries like audio processing librosa and pydub. Achieved 80% accuracy on GTZAN dataset.
- Created a model for captioning and predicting attributes of clothing items based on input images.
- Utilized LSTMs for caption generation and used Tensorflow for attribute prediction and modeling pipeline.
Awards
- Was granted a scholarship of ₹10,000 by BITS for exceptional performance in PS-1 Summer Internship Program.
Volunteering
- Introduced Acapella culture at BITS Goa by leading over 40 vocalists to multiple stage performances and inter college competitions.
- Competed in InBloom’20, Christ University - Bangalore and bagged the 2nd and 3rd position in Western Acapella Competition.