Designed and deployed reasoning-driven AI agents for automated financial analysis, integrating multi-agent orchestration, data intelligence, and enterprise-grade LLM systems.

Designed evaluation methodologies and continuous evaluation pipelines to measure reasoning-agent reliability, robustness, and task performance.

Developed observability and LLMOps infrastructure supporting reproducible evaluation, debugging, and deployment of AI systems.

Developed a real-time financial news analytics system that detects and tracks emerging market trends using transformer-based embeddings and advanced topic modelling.

Designed systematic hyperparameter experiments for BERTopic and evaluated topic quality using coherence and silhouette metrics.

Developed automated monitoring to detect topic drift and changes in model behavior over time.

Pioneered the integration of large language models (LLMs) into robotic systems to advance autonomy, adaptability, and intelligent control.

Fine-tuned code generation models for Robot Operating System (ROS) applications, implemented parameter-efficient fine-tuning approaches and optimized inference pipelines.

Designed cloud-native MLOps infrastructure on AWS for deployment and monitoring.

Conducted research on large language model (LLM) pre-training, focusing on how different neural architectures capture long-distance dependencies (LDDs).

Pre-trained multiple language models from scratch using different neural architectures to study architectural properties and scaling behavior.

Proposed the Long-Short Dependence Ratio (LSDR), a metric for quantifying linguistic dataset complexity and informing model design.

Designed and deployed the web application architecture of a doctor’s appointment system on Amazon AWS, and advised on its Android and iOS apps.

Designed a differential diagnosis system for endocrine diseases using Bayes’ theorem, refining symptom probabilities with medical experts.

Designed and manufactured IEEE 802.15.4 / 6LoWPAN wireless sensor nodes and a border router for home automation, running Contiki OS.

Built an MQTT-based protocol connecting the nodes to AWS IoT Core, and an Android app to monitor sensors and control switches.

Developed and maintained the Autonomous Underwater Vehicle AUV-MAYA and the Autonomous Vertical Profiler (AVP), and took part in their field trials.

Built a Hardware-In-Loop (HIL) simulator, a Google Earth mission planner, mission-control GUIs, and a satellite communication portal for remotely deployed profilers.