NAGALLASATISH

High latency and prediction inaccuracy in high-frequency directional price betting markets.

Built a real-time BTC trading engine for Polymarket using live Binance WebSocket price data and short-term market prediction models. Developed a backend system with live orderbook tracking, WebSocket communication, risk management, and replay/backtesting features using Python and aiohttp. Used XGBoost, LightGBM, Random Forest, and Logistic Regression models to predict BTC price movement in 5-minute markets, achieving AUC scores up to 0.9501. Implemented automated trade execution, live data processing, and real-time monitoring with dashboard and safety controls.

Delayed, server-side comment moderation on social media platforms leading to cyberbullying exposure.

Developed a real-time hate speech detection system for social media platforms such as Twitter, YouTube and Instagram. Implemented text preprocessing, TF-IDF vectorization, and machine learning models to classify content as Hate, Offensive, or Neutral with high accuracy. Integrated the model with a FastAPI backend and a Chrome Extension for real-time detection and inline labeling of comments and posts.

High settlement costs and lack of secure automated routing for merchant payments on-chain.

Developed a blockchain-based payment infrastructure on the Pharos Network enabling secure token transfers, merchant payments, and wallet-to-wallet transactions. Implemented smart contracts, transaction routing, and automated settlement mechanisms to simplify on-chain payments.

Centralized execution, high fees, and lack of transparency in standard betting platforms.

Built a decentralized prediction market platform on the Pharos ecosystem. Developed smart contracts for market creation, trading, settlement, and token-based participation while integrating a modern web interface.

Difficult and slow verification of wallet risk signals and reputation scores for Web3 users.

Built and deployed Base Pulse, a read-only on-chain analytics mini app that analyzes EVM wallet activity on the Base chain. Generates wallet reputation scores and detects strong compromise risk patterns using transaction history and behavior signals. Deployed as a Mini App on Farcaster and Base App with a fast, clean UI for public use.

Robotic, monotonal, and detached synthetic speech outputs in client applications.

Developed an emotion-aware AI speech generation system that enhances traditional text-to-speech applications with emotional intelligence. Built a transformer-based NLP pipeline to detect emotions such as happy, sad, angry, and neutral from user text and dynamically adjust speech parameters including tone, speed, and volume. The system improves listener engagement by generating more natural and context-aware voice output.

Manual, time-consuming storyboard creation for text pitch presentations.

Built an AI-powered storytelling platform that converts text-based ideas and sales pitches into visual storyboards. Implemented scene segmentation, prompt enhancement, AI image generation, PowerPoint export, and video generation workflows. Designed prompt engineering pipelines to improve image quality and generate professional presentations automatically from plain text input.

Black-box understanding of transformer and large language model attention mechanisms.

Implemented the Scaled Dot-Product Attention mechanism from scratch using NumPy to understand the core architecture behind Transformer models and Large Language Models (LLMs). Built the complete attention pipeline including token embeddings, Query-Key-Value (QKV) generation, attention score computation, scaling, masking, softmax normalization, and output tensor generation. Developed attention heatmap visualizations using Matplotlib to analyze word-to-word relationships and model focus patterns, providing an in-depth understanding of how modern Transformer architectures process language.