Resume

Co-Founder, Chief Technology Officer

Stealth AI Startup

• Led a team of 5 in developing tech infrastructure, an AI-driven search, recommendation analytics, and an intelligent document processing engine.
• Drove expansion to 50+ enterprise clients by integrating customer feedback, refining AI models, and ensuring product-market fit.

Project Lead, Machine Learning Initiatives

Branch International

• Led the development of several credit risk models for underwriting that unlocked $100M+ in responsible financial access across India and Africa, enabling data-driven lending for 50M+ underserved users.
• Strengthened AI-driven lending systems through reliable training and inference workflows, supported by cross-functional collaboration and multiple initiatives that elevated model quality and deployment standards.

Machine Learning Engineer

Branch International

• Drove the development of a neural network-based credit risk model from unstructured SMS data, ensuring transparent and interpretable underwriting decisions aligned with regulatory standards.
• Accelerated feature fetch processing by 8 times through a distributed head-worker system using Ray, fine-tuning service worker and threadpool configurations, optimizing SQL queries, and intelligent batching of I/O- & CPU- intensive requests. (https://cosmic-heart.github.io/post/2025/feature-fetch-optimization/)
• Developed a credit model using bureau data for the premium user segment, achieving a 3% delinquency rate, at 29% APR reduction.
• Built a fully automated, scalable training infrastructure on AWS, reducing model training time and costs by fivefold.
• Integrated a RabbitMQ-based priority queue system to orchestrate reliable communication between backend and AI services.
• Led the effort of integrating bureau and account aggregator data sources into our data & feature engineering pipeline
• Decreased feature service latency by threefold by rewriting critical CPU-intensive components in Rust
• Refined training sample queries and incorporated shorter window targets, enabling the use of recent data along with improving data confidence & generalizability across all user segments.

Machine Learning Engineer

Branch International

• Trained robust BERT-based cashflow classification & extraction SMS model with AUC of 0.99, increasing data coverage threefold, and maintained year-long operational stability.
• Engineered bureau-based features and trained a credit model for first-time borrowers, resulting in a 6% AUC improvement and enabling $90 million in financial access for underserved users.
• Developed a parallel processing algorithm to identify issues in retrospective bureau data and transform it into Credit Reports at 100K+ files/min.

Engineering Intern

VIT University

• Built a parallel and distributed computing workspace using Ray, NFS to accelerate ML training and inference.
• Configured Linux & ML Frameworks on IBM PowerPC ppc64le architecture.
• Architected and deployed an AI Learning Platform on distributed clusters to support hands-on learning for junior students.

Machine Learning Engineer

Gutsy Innovation

• Built a real-time face detection & recognition system using YOLOv7 and ArcFace.
• Integrated Kalman filter–based face tracking and ESPCN super resolution to enhance image clarity.
• Deployed this AI surveillance service in edge devices using DeepStream SDK and integrated it with Azure IoT Hub for monitoring.

B.Tech in Computer Science & Engineering, Specialization in Artificial Intelligence & Machine Learning

VIT University

GPA: 8.92/10

Relevant Coursework: Machine Learning, Deep Learning, Natural Language Processing, Data Structures & Algorithms, Operating System, Networks & Security, DBMS, Linear Algebra, Statistics, Software Engineering, Product Development & Entrepreneurship

Instruction-Aligned AI Tutor

• Fine-tuned LLaMA-7B using RLHF on synthetic Q&A corpora with adversarial prompts, improving pedagogical alignment and mitigating prompt injection vulnerabilities.
• Engineered a high-throughput LLM-RAG system using vLLM and Ray, integrated with Qdrant and LangChain for contextual, real-time doubt clarification.
• Built a scalable Django backend with PostgreSQL and Cassandra to retain user interactions for continuous retraining and future model iterations.

Multimodal Malware Forensics

• Trained a CoAtNet Transformer on 2M malware binaries converted into dual-channel images using byteplot and bigram-DCT transforms to learn morphological malware signatures.
• Trained LightGBM models on static PE structural attributes extracted from binaries, and built an LSTM classifier over dynamic system-call traces obtained from sandbox execution logs.
• Ensembled all classifiers using a meta-model for production inference, and developed a cross-platform Rust-based client that uploads files to an AI-driven backend with real-time model scoring visualized via Azure Power BI.

Spatiotemporal Anomaly Perception

• Trained a SwinTransformer as a variational autoencoder for frame reconstruction, and fused its encoder with LSTM decoders to capture temporal anomalies in UCF-Crime surveillance streams.
• Optimized anomaly classification by adding an SVM-based prefilter, calibrated to capture minute anomalies, which is then verified by an LSTM network to reduce computation and false positives.
• Deployed an edge-ready WebSocket pipeline for multi-camera ingestion and low-latency anomaly alerts.

Pix2Struct-Based Vision-Language Graph Parsing

• Performed visual reasoning on charts and plots to make them accessible for visually impaired users.
• Used EfficientNetV2 for chart classification and applied Matcha (Pix2Struct) for bar, line, and dot plot interpretation, along with Faster R-CNN for scatter plot derendering.

Hall of Fame Award for Best Performing Engineer

Branch International

Winner

Branch Hackathon – For Building a GenAI Credit Risk Model

Branch International

Runner

Best Final Year Project, VIT University

VIT University

Winner, for developing an on-premise AI learning platform

Branch Hackathon – For developing an AI Agent to assist engineers with documentation

Branch International

Winner

Jee Mains

All India

98.92%ile, conducted all over India

10th Grade Board Exams

Board

Top 1%

Public Relations Officer

Rotaract Club, VIT University

• Spearheaded communication, outreach, and marketing efforts for the club.
• Organized fundraising events, coordinated donation drives for underprivileged communities,
• Played a key role in the execution of events such as blood donation camps, cleanliness drives, and educational workshops.