projects.py

import streamlit as st

def display_projects():
    st.title('My Projects')

    # Define tab titles
    tab_titles = [
        "Semantic Video Search",
        "Video-QA System",
        "AdHubby: AI-Powered Marketing Campaign Generator",
        "Resume & CV Crafter",
        "Multi-Agent Job Search",
        "Resume Easz",
        "Job Easz",
        "Bitcoin Lightning Optimization",
        "National Infrastructure Monitoring",
        "Stock Market Analysis",
        "Twitter Trend Analysis",
        "Restaurant Recommendation",
        "ASL Translator",
        "Squat Easy"
    ]

    # Create tabs
    tabs = st.tabs(tab_titles)

    # Add content to each tab
    with tabs[0]:
        st.subheader("🏆 Semantic Video Search")
        st.markdown("**Winner - Liquid AI Hackathon** | Nov 2025")
        st.markdown("""
        - **Description**: Multimodal app for AI-driven video analysis and semantic search using LiquidAI's LFM2 models
        - **Key Features**:
            • Frame-by-frame analysis with LFM2-VL-450M
            • Smart filtering to skip redundant frames
            • Semantic search with transformer embeddings
            • Clip extraction around matched frames
            • YouTube URL + video upload support
        - **Technical Implementation**:
            • Temporal context modeling (t-1, t, t+1)
            • 384-d sentence transformer embeddings (MiniLM-L6-v2)
            • Top-K retrieval with cosine similarity
            • Real-time progress tracking
        - **Technologies**: LFM2-VL-450M, MiniLM-L6-v2, PyTorch, OpenCV, Streamlit, yt-dlp
        - **Skills**: Computer Vision, Vision-Language Models, Semantic Search
        """)

    # Video-QA System
    with tabs[1]:
        st.subheader("Video-QA System")
        st.markdown("**E2B + Groq + MCP + Notion** | Nov 2025")
        st.markdown("""
        - **Description**: Video Question-Answering system that processes videos in E2B sandbox, generates frame summaries via Groq VLM, and stores knowledge in Notion through MCP
        - **Key Features**:
            • Frame extraction + VLM analysis in isolated E2B VM
            • MCP integration for structured video knowledge storage
            • Chat interface for context-based Q&A
            • Timestamp-accurate answers from video context
            • YouTube links + local upload support
        - **Motivation**: Solve difficulty of revisiting complex YouTube walkthroughs—turning videos into searchable, chat-ready knowledge bases
        - **Technologies**: E2B, Groq API, Model Context Protocol (MCP), Notion API, VLM
        - **Demo**: [YouTube Demo](https://youtu.be/u08LN_Gh4HE)
        - **Reference**: [GitHub](https://github.com/niharpalem/VideoQA)
        """)

    with tabs[2]:
        st.subheader("AdHubby: AI-Powered Marketing Campaign Generator")
        st.markdown("""
        - **Description**: Multi-agent multimodal AI system that creates comprehensive marketing campaigns by combining strategic analysis, visual generation, and local landmark integration for small businesses
        - **Key Features**:
            • Smart Marketing Intelligence with AI-powered briefing generation and competitive analysis
            • 9 Professional Marketing Formats (social carousels, billboards, email newsletters, print brochures, etc.)
            • Local Integration automatically incorporating 5 relevant landmarks into marketing visuals
            • Visual Style Engine with 5 artistic styles (Anime, Cartoon, Minimalist, Watercolor, Vintage)
        - **Technical Achievements**:
            • Engineered multi-agent architecture: Strategic Agent (Tenstorrent-powered Llama-3.1-8B accessed through Koyeb's cloud platform), Visual Agent (Stable Diffusion XL), QA Agent (LLM-as-judge)
            • Implemented advanced prompting with domain context injection and two-stage LLM judging system
            • Built comprehensive workflow delivering ready-to-deploy marketing materials in under 8 hours (hackathon timeframe)
            • Developed innovative geo-targeted visual asset generation for location-based marketing
        - **Technologies**: Streamlit, Tenstorrent (via Koyeb), Llama-3.1-8B, Stable Diffusion XL, Hugging Face API, Multi-Agent Architecture
        - **Achievement**: 🏆 **Tenstorrent Multi-Agent Hackathon Recognition** - Built in **5** hours
        - **Reference**: [GitHub Repository](https://github.com/niharpalem/AdHubby.com)
        """)
    with tabs[3]:
        st.subheader("LLM-powered Resume & CV Crafter")
        st.markdown("""
        - **Description**: Developed AI platform combining LLaMA-3 70B and Deepseek R1 with low-temperature settings for stable, tailored resume and CV generation
        - **Key Features**:
            • Smart Matching Algorithm analyzing profiles against job requirements
            • LaTeX-Powered Resumes with professional formatting
            • Automated 4-paragraph Cover Letter Generation
            • Performance Metrics evaluating match quality
        - **Technical Achievements**:
            • Implemented dual-agent architecture: LLaMA-3 8B for profile analysis and 70B for LaTeX generation
            • Engineered JSON schema validation system for error-free template integration
            • Achieved 5,000+ LinkedIn impressions with 80% reduction in creation time
        - **Technologies**: Streamlit, GROQ API (LLaMA-3 70B), LaTeX, JSON Schema
        - **Reference**: [Link to Project](https://huggingface.co/spaces/Niharmahesh/Resume_and_CV_crafter)
        """)

    with tabs[4]:
        st.subheader("Multi-Agent Job Search System")
        st.markdown("""
        - **Description**: Built an AI-powered job search assistant using dual-LLaMA architecture for comprehensive job matching and analysis
        - **Key Features**:
            • Real-time scraping across LinkedIn, Glassdoor, Indeed, ZipRecruiter
            • Advanced resume parsing and job matching
            • Intelligent compatibility scoring system
        - **Technical Achievements**:
            • Developed batch processing pipeline handling 60+ positions/search
            • Reduced job search time by 80% through accurate matching
            • Implemented specialized agents for input processing, scraping, and analysis
        - **Technologies**: GROQ API, jobspy, Streamlit, Pandas, LLMOps
        - **Reference**: [Link to Project](https://huggingface.co/spaces/Niharmahesh/Multi_Agent_Job_search_and_match)
        """)

    with tabs[5]:
        st.subheader("Resume Easz")
        st.markdown("""
        - **Description**: Created an AI-driven resume analysis and enhancement tool using LLaMA 3.3 model
        - **Key Features**:
            • Quick and in-depth resume analysis options
            • Comprehensive skill gap analysis
            • ATS compatibility optimization
            • Multiple output formats (DOCX, HTML, TXT)
        - **Technical Implementation**:
            • Integrated GROQ API for advanced language processing
            • Built visual diff system for resume changes
            • Developed custom prompt engineering pipeline
        - **Technologies**: GROQ API, Streamlit, Python, LLM
        - **Reference**: [Link to Project](https://resume-easz.streamlit.app/)
        """)

    with tabs[6]:
        st.subheader("Job Easz")
        st.markdown("""
        - **Description**: Engineered comprehensive job aggregation platform for data roles with advanced analytics
        - **Technical Achievements**:
            • Designed Airflow pipeline with exponential backoff retry (120-480s intervals)
            • Optimized concurrent processing reducing runtime from 2h to 40min
            • Processes ~3000 daily job listings across various data roles
        - **Key Features**:
            • Daily updates with comprehensive job role coverage
            • Custom filtering by role and location
            • Interactive dashboard for market trends
            • Automated ETL pipeline
        - **Technologies**: Python, Airflow, ThreadPoolExecutor, Hugging Face Datasets
        - **Reference**: [Link to Project](https://huggingface.co/spaces/Niharmahesh/job_easz)
        """)

    with tabs[7]:
        st.subheader("Bitcoin Lightning Path Optimization")
        st.markdown("""
        - **Description**: Advanced payment routing optimization system for Bitcoin Lightning Network
        - **Technical Achievements**:
            • Developed ML classifiers achieving 98.77-99.10% accuracy
            • Implemented tri-model consensus system for optimal routing
            • Engineered ensemble models with 0.98 F1-scores
        - **Implementation Details**:
            • Created simulation environment for multi-channel transactions
            • Optimized graph-based algorithms for payment routing
            • Integrated with Lightning payment interceptor
        - **Technologies**: XGBoost, Random Forest, AdaBoost, Graph Algorithms
        """)

    with tabs[8]:
        st.subheader("National Infrastructure Monitoring")
        st.markdown("""
        - **Description**: Developed satellite imagery analysis system for infrastructure change detection
        - **Technical Achievements**:
            • Fine-tuned ViT+ResNet-101 ensemble on 40GB satellite dataset
            • Achieved 85% accuracy in change detection
            • Implemented 8 parallel GPU threads for enhanced performance
        - **Key Features**:
            • Temporal analysis with 1km resolution
            • Interactive map interface with bounding box selection
            • Automatic image chipping for 256x256 inputs
            • Contrast adjustment optimization
        - **Technologies**: Change ViT Model, Google Earth Engine, PyTorch, Computer Vision
        - **Reference**: [Link to Project](https://huggingface.co/spaces/Niharmahesh/Data298)
        """)

    with tabs[9]:
        st.subheader("Stock Market Analysis with OpenAI Integration")
        st.markdown("""
        - **Description**: Created comprehensive stock market analysis system with multilingual capabilities
        - **Technical Achievements**:
            • Built Spark streaming pipeline with 30% efficiency improvement
            • Orchestrated Airflow Docker pipeline for Snowflake integration
            • Developed bilingual GPT-3.5 chatbot for SQL query generation
        - **Key Features**:
            • Real-time financial metric calculations
            • Custom indicator generation
            • Multilingual query support
            • Automated data warehousing
        - **Technologies**: PySpark, Apache Airflow, Snowflake, OpenAI GPT-3.5
        """)

    with tabs[10]:
        st.subheader("Twitter Trend Analysis")
        st.markdown("""
        - **Description**: Engineered comprehensive Twitter analytics platform using GCP services
        - **Technical Achievements**:
            • Developed GCP pipeline processing 40k tweets
            • Achieved 40% efficiency improvement through custom Airflow operators
            • Implemented real-time trend analysis algorithms
        - **Key Features**:
            • Automated ETL workflows
            • Interactive Tableau dashboards
            • Viral metrics tracking
            • Engagement rate calculations
        - **Technologies**: Google Cloud Platform, BigQuery, Apache Airflow, Tableau
        """)

    with tabs[11]:
        st.subheader("Restaurant Recommendation System")
        st.markdown("""
        - **Description**: Built hybrid recommendation system combining multiple filtering approaches
        - **Technical Achievements**:
            • Created hybrid TF-IDF and SVD-based filtering system
            • Achieved 43% improvement in recommendation relevance
            • Reduced computation time by 65%
        - **Key Features**:
            • Location-based suggestions
            • Personalized recommendations
            • Interactive web interface
            • Efficient matrix factorization
        - **Technologies**: Collaborative Filtering, Content-Based Filtering, Flask, Folium
        """)

    with tabs[12]:
        st.subheader("ASL Translator")
        st.markdown("""
        - **Description**: Developed real-time American Sign Language translation system
        - **Technical Achievements**:
            • Achieved 95% accuracy in real-time gesture interpretation
            • Implemented adaptive hand skeleton GIF generator
            • Optimized MediaPipe integration for point detection
        - **Key Features**:
            • Real-time hand tracking
            • Visual feedback system
            • Intuitive gesture recognition
            • Accessible interface
        - **Technologies**: MediaPipe Hand Detection, Random Forest, Hugging Face Platform
        - **Reference**: [Link to Project](https://huggingface.co/spaces/Niharmahesh/slr-easz)
        """)

    with tabs[13]:
        st.subheader("Squat Easy")
        st.markdown("""
        - **Description**: Developed deep learning system for squat form analysis and error detection
        - **Technical Achievements**:
            • Engineered custom BiLSTM architecture in PyTorch
            • Achieved 81% training and 75% test accuracy
            • Implemented CUDA-based GPU acceleration
        - **Key Features**:
            • Real-time form analysis
            • Six-type error classification
            • Video processing pipeline
            • Performance optimization
        - **Technologies**: PyTorch, BiLSTM, CUDA, Object-Oriented Programming
        - **Reference**: [Link to Project](https://github.com/niharpalem/squateasy_DL)
        """)