Emerging Trend Detection using Reinforcement Learning A2C
Advantage Actor-Critic (A2C) for emerging trends in social posts: the agent optimizes actions (post, edit, delete) using engagement signals such as likes, retweets, and quotes.
View Project →Real-Time Hand Gesture Recognition using Vision Transformer (ViT)
Vision Transformer (ViT) for live webcam gesture recognition, trained on the Chinese Hand Gesture Number dataset and ASL with real-time video processing and an interactive UI.
View Project →Deep Q-Network (DQN) Variants Analysis for Optimization
PyTorch implementations of Classic DQN, Double DQN, and Dueling DQN on OpenAI Gymnasium with a comparative analysis framework and performance tuning.
View Project →NanoGPT Fine-Tune for Domain-Specific Tasks
Fine-tuned NanoGPT for quotes and poems: dataset prep, training, and evaluation for small-domain text generation.
View Project →Context AI Translator
Select-to-translate with shortcuts and context-aware examples across many languages, difficulty levels for example sentences, optional auto-translate on selection, flashcards, and privacy-focused API key settings.
View Project →Speech-to-Speech AI Agent (Echo Mortgage Voice UI)
TypeScript speech-to-speech pipeline with real-time voice UX; deployed experience for mortgage-rate consultation, calculators, and loan-type flows (Conventional, FHA, VA, Jumbo, ARM).
View Project →HR AI Agent
Voice-oriented AI agent frontend patterns: transcription, playback controls, agent state visualization, and Tavus-style avatar-oriented study flows.
View Project →Talent Recruiter
TypeScript-heavy client for job discovery and recruiting workflows: structured templates, styling system, and integrated JS behavior.
View Project →WebSpider
Python tooling for job-related scraping: automated browser interaction, external APIs, search over collected data, and analysis pipelines.
View Project →Voice Agent Python (LiveKit)
LiveKit-based voice assistant: Deepgram STT, Cartesia TTS, GPT-4, turn detection, noise handling, tests, metrics, and Docker-ready deployment for web, mobile, or telephony.
View Project →Conservative Q-Learning (CQL) & Behavior Cloning
RL homework codebase: Behavior Cloning and Conservative Q-Learning on offline state–action data with experiments for comparison and hyperparameter sweeps.
View Project →Top Studies Recommender
Python study-recommendation prototype using classical ML classification on curated study data.
View Project →ViT–RFC Fusion for Real-Time Hand Gesture Recognition
Alternate ViT + RFC fusion implementation for interactive real-time gesture recognition with webcam and multi-dataset training.
View Project →
.png){kind=logo}
