Welcome to my portfolio
Akshat Gurbuxani
AI Researcher @ Semiotic Labs
Building intelligent systems that transform complex challenges into scalable AI solutions
About Me
Technologies
Here are a few technologies I've been working with recently:
I work on turning research ideas into working AI systems. Over the last two years, that's meant training large language models, wiring them into distributed systems, and making sure they hold up in production.
Most of my work lives end-to-end: taking an idea from research or theory, turning it into a model, and deploying it at scale. Along the way, I've worked on multimodal learning, multi-agent systems, and real-time data pipelines.
I care about the gap between what papers promise and what systems actually do, and I spend most of my time trying to close it.
Experience
- Trained SOTA multi-modal NFT spam detection system across 8 blockchain networks using fine-tuned LLM, GNN, XGBoost with Mixture-of-Experts ensemble, achieving 87% accuracy, correctly flagging 150,000+ malicious contracts
- Built trajectory-aware multi-agent pipeline with DSPy ReAct agents, integrating The Graph's Token API and Subgraphs with judge-agent for evaluation; improved response accuracy by 70%, reduced query latency by 60% in production
- Designed RAG-based chatbot with BPS policy repository, automating document checks and reducing manual effort by 85%
- Aligned FAISS with vector and lexical search, metadata-driven clustering, and a multi-agent LLM pipeline, enhancing retrieval speed by 70% and document relevance by 90%
Machine Learning Engineer
Optum Global Solutions Pvt Ltd
July 2021 – Aug. 2023
Hyderabad, India
- Productionized neural collaborative filtering model to personalize insurance policy recommendations, increasing retention by 70% and page traffic by 86%. Integrated Kafka and RESTful APIs for real-time content delivery
- Optimized ML pipelines using Apache Spark, reducing model training time by 75%, enabling faster model iterations and scalable deployments
- Developed LSTM-based time-series forecasting model for website traffic prediction, attaining 80% accuracy. Automated testing workflows with pytest, Docker, CI/CD, Jenkins, and Kubernetes reducing manual testing by 70%
- Engineered distributed data pipeline using Apache Kafka and Apache Flink for real-time data processing, reducing data latency from 40 minutes to 5 minutes, and increasing data throughput by 80%
- Structured multi-node model deployment architecture using Kubernetes and Docker, enabling parallel inference across multiple GPUs, reducing inference time by 70%, supporting high-throughput applications
Projects
A collection of my work showcasing various technologies and problem-solving approaches
AI Mathematical Olympiad
April 2024 – June 2024
Secured silver medal in Kaggle's AIMO competition by applying advanced LLM techniques to solve olympiad-level math problems.
AI Stock Trader
Jan. 2024 – May 2024
LLM-based trading pipeline leveraging real-time news and historical data to automate E-mini S&P 500 futures trades.
Itinerary Planner
Sep. 2024 – Nov. 2024
AI-powered trip planner generating personalized itineraries based on dates, budget, and preferences.
3D Human Pose Estimation
April 2024 – May 2024
Spatiotemporal Transformer and LSTM models to transform 2D to 3D poses using monocular RGB videos.
Simple UniverSeg
Feb. 2024 – April 2024
U-net with CrossBlock and few-shot learning for medical image segmentation without re-training.
Education
Boston University
Master of Science in Artificial Intelligence
Sep. 2023 – Dec. 2024
Boston, MA
Activities
- • Graduate Teaching Assistant: Grad Intro to Natural Language Processing (NLP)
- • Graduate Teaching Assistant: Data Science Tools and Applications
Related Coursework
SRM Institute of Science and Technology
Bachelor of Science in Computer Science
Aug. 2017 – May 2021
Chennai, India