Analyst Data Science at Sirius AI (2025-05 – Present)
- Engineered a privacy-first AI Meeting Assistant by optimizing Whisper and Llama for local edge execution, implemented RAG pipelines to deliver real-time context with zero cloud data exposure.
- Built a high-volume email automation ecosystem using Azure Function Apps, Service Bus, and Azure AI Document Intelligence for multi-modal PDF understanding and enterprise routing.
- Developed a tiered AI classification engine using Qwen embeddings for XGBoost and fine-tuned Gemma, incorporating Naïve Bayes for Arabic detection and Triton Inference Server for high-throughput deployment.
- Set up autonomous AI agent via AutoGen and Semantic Kernel for end-to-end processing, transforming raw email queries into structured case creations and enhancing model modality with synthetic data.
Computer Vision Research Engineer at Xterra Robotics (2024-12 – 2025-05)
- Optimized vision models YOLO, RF-DETR, CNN/transformers for object detection and real-time semantic segmentation, boosting mAP and enabling low-latency robot path planning.
- Implemented 3D point cloud segmentation pipelines (voxelization, clustering, feature extraction) to enhance SLAM-based mapping and autonomous navigation.
- Quantized large language and vision models for edge deployment on Jetson Orin, reducing compute and memory costs while maintaining performance.
Applied AI Researcher at Spyne.AI (2024-06 – 2024-12)
- Built deep learning models - MVANet, BiRefNet for dichotomous segmentation on automotive features, improving accuracy by 15% through augmentation and hyperparameter tuning.
- Developed image processing pipelines (OpenCV) with resizing, normalization, and enhancement, creating dynamic overlays that improved analysis across 1000+ images.
- Deployed models on NVIDIA Triton Server, enabling real-time, low-latency segmentation with ensembles and async inference, handling 10K+ images per day.
Data Science Intern at Router Protocol (2023-03 – 2023-05)
- Built web scraping pipelines with Python (Scrapy, Selenium) and Node.js, extracting 100K+ crypto data points from CMC and ICOdrops.
- Built NLP and sentiment analysis workflows using spaCy, NLTK, VADER, and TextBlob, processing 10K+ Twitter comments to model community sentiment and token perception.
Machine Learning Intern at Treacle Technologies Pvt. Ltd (2022-04 – 2022-06)
- Developed full-stack web interface from scratch using HTML, CSS, JavaScript, React.js, and AngularJS, including dashboards to visualize cybersecurity metrics like DDoS frequency.
- Implemented ML models for DDoS and botnet attack prediction, applying algorithms like K-Means and Random Forest, with SMOTE/ADASYN to handle dataset imbalance.
