3. Data, Science & AI

Architects and troubleshoots AI agents using AWS Strands, RAG pipelines, and AgentCore for communication patterns and performance optimization.

Framework for building LLM applications using model wrappers, prompt templates, and chaining techniques.

Optimizes prompts for LLMs and AI agents, specializing in cognitive architecture, persona caching, and agent-to-agent communication.

Provides tools for Direct Preference Optimization (DPO) training, including dataset preparation, implicit reward modeling, and beta tuning for stable language model training without explicit reward models.

Fine-tunes vision models using FastVisionModel, supporting Pixtral, Ministral VL, and vision-specific LoRA configurations with image-text datasets.

Generates Google Ads responsive search ad copy using AI and embeddings, optimized for performance and compliance with platform constraints.

Integrates Scorable LLM-as-a-Judge for evaluating, monitoring, and safeguarding LLM application outputs in real-time.

ComfyUI provides a node-based interface for Stable Diffusion, enabling GPU-accelerated image generation with custom nodes and CivitAI model integration.

Conducts advanced QLoRA fine-tuning experiments including quantization, rank selection, and multi-adapter strategies.

Creates high-quality scientific figures for SCI journal papers by analyzing project code and generating data visualizations.

Enables building AI agents, RAG applications, and prompt flows via Azure AI Foundry for orchestrating AI solutions on Azure.

Creates interactive data visualization dashboards and KPI displays using Recharts for analytics purposes.

Enables building AI agents on AWS Bedrock AgentCore with tool integration, orchestration, and Bedrock model interaction.

Comprehensive ACM system for predictive maintenance, including RUL forecasting, detector tuning, SQL data management, and observability for industrial equipment monitoring.

Generates high-quality images with text rendering and character consistency using Google Gemini 3 Pro Image API for professional visual content.

Optimizes TensorFlow.js WebGPU backend and WebLLM for efficient offline inference with memory safety in heavy machine learning models.

Sets actionable goals for autonomous agents using intent analysis and SMART criteria, ensuring tasks are specific, measurable, achievable, relevant, and time-bound.

Adjusts model decision thresholds to apply high-confidence decisions and enforce safety margins for low-confidence predictions, enhancing reliability in AI systems.

Classifies skin lesions including Melanoma and Basal Cell Carcinoma using a pre-trained MobileNetV3 model in TensorFlow.js.

Extracts vector embeddings for fairness analysis using MobileNetV3 and FairDisCo disentanglement.

Orchestrates goal-oriented action planning for clinical data analysis pipelines, enabling efficient execution of healthcare data processing tasks.

Normalizes image histograms to preserve melanin features using Canvas API, preparing images for medical or dermatological analysis.

Builds model-agnostic AI agents, ReAct patterns, and multi-agent systems using Strands Agents SDK on AWS for production deployment.

Automatically generates SCI-compliant research papers from project code and problem-solving processes.