Jun 2025 – Present · 9 mos · California, United States · Hybrid

• Working under the guidance of Prof. Solomon Berhe, Professor at the University of the Pacific, on research focused on sentiment analysis of Reddit post titles.
• Developing end-to-end sentiment trajectory analysis systems to examine how author sentiment compares to community responses in Reddit discussions about software updates.
• Implementing automated sentiment analysis pipelines using VADER (Valence Aware Dictionary and sEntiment Reasoner) for large-scale Reddit post and comment analysis.
• Fetching and processing real-time data from Reddit API endpoints, collecting posts with metadata including titles, authors, comments, scores, and timestamps.
• Building quality filtering systems with minimum thresholds for author replies (3+) and community comments (5+) to ensure trajectory reliability.
• Creating trajectory reliability metrics to assess consistency and trend strength of sentiment patterns across author replies and community responses.
• Developing quality scoring algorithms combining author engagement, community participation, and trajectory reliability for optimal post selection.
• Implementing automated categorization systems classifying content into positive, negative, and neutral categories based on compound sentiment scores.
• Building Python scripts for data filtering, sentiment analysis, and visualization generation with comprehensive error handling and logging.
• Generating matplotlib visualizations showing sentiment trajectories with separate plots for positive, negative, and neutral categories.
• Designing interactive web-based dashboards using HTML, CSS, JavaScript, and Chart.js for dynamic data exploration.
• Creating comprehensive reference documentation mapping visualized posts to original Reddit URLs with detailed metadata and quality metrics.
• Building scalable data processing pipelines analyzing 324+ quality-filtered posts from thousands of initial Reddit entries with reproducible workflows.

Jan 2024 – May 2024 · 4 mos · California, United States · On-site

• Developing LIMFADD (LLM-enabled Instagram Multi-Class Fake Account Detection Dataset), the first dataset to classify Instagram accounts into four categories: Real, Spam, Scam, and Bot.
• Collecting and curating real Instagram profile data, including follower/following ratios, post activity, profile metadata, and engagement behavior for baseline modeling.
• Manually identifying spam accounts by analyzing repetitive comment patterns, automated messaging behavior, and abnormal engagement spikes across posts.
• Identifying scam accounts based on phishing patterns, suspicious URLs, fake giveaways, impersonation behavior, and fraudulent promotional tactics.
• Categorizing bot accounts by detecting automated behavior patterns such as mass-following, generic comments, and abnormal posting frequency.
• Constructing a balanced base dataset with equal representation across all four labels to avoid class imbalance during training.
• Expanding dataset size via LLM-based synthetic data generation using ChatGPT to extrapolate realistic feature distributions while preserving class-specific behavior patterns.
• Performing data preprocessing, including cleaning, normalization, error handling (e.g, follower/following divide-by-zero cases), and class-label encoding for machine learning.
• Training a Deep Neural Network (DNN) using TensorFlow/Keras with ReLU activations and softmax output for multi-class classification.
• Achieving overall Accuracy, Macro-Precision, Macro-Recall, and Macro-F1 scores of 0.97, outperforming state-of-the-art binary fake account datasets.
• Conducting XAI analysis using LIME to interpret model predictions and identify influential behavioral features such as followers, posts, threads activity, and following patterns.
• Comparing LIMFADD against benchmark Instagram datasets and demonstrating significantly higher performance in both binary and multi-class settings.
• Contributing to long-term efforts to enhance security and fraud detection on large social media platforms.

Jun 2025 – Present · 9 mos · California, United States

• Co‑Founder & Technical Lead – Stock Crusher (AI Stock Intelligence Platform)
• Built an AI‑powered stock analysis platform that turns Reddit, Twitter/X, news, Yahoo Finance, and Finnhub data into clear BUY/SELL/HOLD calls with confidence and risk levels.
• Owned the full stack – designed the architecture, wrote the Python/Flask backend and APIs, and built a modern, mobile‑first UI using Tailwind CSS.
• Implemented a dual‑AI engine combining Google Gemini and Perplexity, with smart prompt design and a consensus algorithm to handle conflicting model outputs.
• Engineered a real‑time data pipeline that continuously pulls, cleans, and normalizes multi‑source market data, with graceful fallbacks for API failures and rate limits.
• Created a proprietary momentum score using changes in sentiment, volume, and influence to highlight stocks with emerging upside or downside trends.
• Logged every analysis historically so users can see past recommendations, momentum rankings, and how the AI’s view on each ticker has evolved over time.
• Focused on explainability and UX by surfacing the key factors behind each recommendation and allowing users to drill into the underlying posts, articles, and metrics.
• Hardened the platform for production with environment‑based configuration, API key validation, robust error handling, and deployment‑ready settings.
• Led product and technical decisions as co‑founder, turning a high‑level idea (“use AI to read the market”) into a concrete roadmap, working product, and scalable architecture.