Skip to content
View harsha4261's full-sized avatar

Highlights

  • Pro

Organizations

@AIHUB-VVIT

Block or report harsha4261

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
harsha4261/README.md

🧠 Who I Am

class HarshaVardhanReddyEmani:
    def __init__(self):
        self.role        = "AI Research & Systems Engineering Student"
        self.location    = "Andhra Pradesh, India"
        self.education   = "B.Tech CSE @ VVITU  β€’  GPA: 8.7 / 10"
        self.live_system = "SocialGPT β€” Production Multi-Agent RAG (5,000+ users)"
        self.url         = "socialgpt.aihub-vvitu.social"

        self.expertise   = [
            "LLM fine-tuning: LoRA, SFTTrainer, DDP on AMD MI300X / Intel Gaudi3 / NVIDIA L40S",
            "Multi-agent orchestration with LangGraph + Claude claude-opus-4-6",
            "Full RAG pipelines: PDF ingestion β†’ HNSW vector search (SurrealDB)",
            "HPC: OpenMP, Berkeley DB, performance profiling, C/C++ graph structures",
            "Production infra: FastAPI, Docker, Prometheus, Grafana, LangSmith",
        ]

        self.currently   = [
            "Maintaining SocialGPT in production (8 specialized sub-agents, RBAC, multilingual)",
            "Exploring mechanistic interpretability & AI safety research",
            "Contributing to AIHUB VVIT β€” mentoring 100+ peers in applied ML",
        ]

        self.languages   = {
            "code":   ["Python", "C", "C++", "TypeScript", "SQL", "Java", "Dart"],
            "spoken": ["English", "Telugu", "Hindi", "Tamil (system support)"],
        }

        self.mission     = "Build AI systems that are powerful, safe, and genuinely useful."

πŸ† Highlights at a Glance

πŸš€ Shipped πŸ”¬ Trained πŸ›οΈ Spoken 🀝 Mentored
SocialGPT in production serving 5,000+ users Medical LLM (7B) via LoRA on 8Γ— AMD MI300X SCI 2025 β€” 3-hr tutorial on LLM Math Foundations 100+ peers at AIHUB VVIT

πŸ› οΈ Tech Stack

Languages

Python C C++ TypeScript JavaScript Java Dart SQL

LLM Β· Agents Β· ML

PyTorch Hugging Face LangChain Anthropic DeepSpeed Scikit-learn LlamaIndex

HPC Β· GPU Platforms

CUDA AMD ROCm Intel Gaudi OpenMP

Databases Β· Vector Search

MongoDB SurrealDB Redis MySQL MinIO

Infrastructure Β· Observability

Docker FastAPI Nginx Prometheus Grafana LangSmith GitHub Actions Linux

Frontend

React Vite Tailwind CSS shadcn/ui


πŸ”₯ Flagship Projects

πŸ€– SocialGPT β€” Production Multi-Agent Platform

Live @ socialgpt.aihub-vvitu.social

LangGraph Claude SurrealDB NVIDIA

  • πŸ›οΈ 5,000+ users β€” students, faculty, admins at VVITU
  • 🧩 LangGraph orchestrator coordinating 8 specialized sub-agents (ERP, RAG, Bus, HireHub, Teams, Web, Analytics, TrackCoders) with parallel tool calls & stateful multi-turn memory
  • πŸ“„ Full RAG over 18 years of institutional PDFs β†’ LlamaParse + Surya OCR β†’ BAAI/bge-m3 (1024-dim, ~18K chunks) β†’ HNSW in SurrealDB
  • πŸ” RBAC enforced at FastAPI middleware + inside every LangChain tool
  • 🌐 Multilingual: English, Telugu, Hindi, Tamil
  • πŸ”­ Full observability: Prometheus + Grafana + Loki + LangSmith per-agent tracing; validated against 42-prompt test suite

🧬 Medical LLM β€” Cross-Hardware Training

Fine-tuning & training across AMD, Intel, NVIDIA GPU platforms

PyTorch AMD ROCm LoRA DDP

  • πŸ”§ Built a GPT-2-style SLM from scratch: Causal Self-Attention, LayerNorm, Transformer blocks β€” deep mechanistic understanding
  • ⚑ Multi-GPU training on 8Γ— AMD MI300X with DDP + gradient accumulation; benchmarked vs. NVIDIA L40S
  • πŸ’Š SFT of 7B medical LLM via LoRA + mixed-precision targeting verifiable, reproducible medical reasoning
  • πŸ“Š Cross-platform throughput & memory efficiency benchmarking: MI300X vs. Gaudi3 vs. L40S

πŸ“Š TrackCoders β€” Campus Analytics Platform

Production-grade coding performance dashboard

FastAPI MongoDB React Docker

  • 🏫 Serving 5,000+ students & faculty campus-wide with real-time dashboards
  • ⚑ Sub-second query latency with role-based data access and scalable MongoDB backend
  • πŸš€ Deployed on Linux with Docker; now integrated as a dedicated sub-agent inside SocialGPT

🎯 Hierarchical Prompt Classification Pipeline

ML pipeline relevant to scalable oversight & prompt robustness

Python Scikit-learn NLP

  • πŸ” TF-IDF vectorization + unsupervised clustering β†’ hierarchical classification
  • πŸ”„ Dataset augmentation for improved generalization across prompt distributions
  • πŸ’Ύ Pickle-based model persistence for offline & web-based inference

πŸ’Ό Experience & Speaking

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚  🎀  Technical Speaker β€” SCI 2025 Conference                         December 2025   β”‚
β”‚      3-hour tutorial: "Mathematical Foundations for Large & Small Language Models"   β”‚
β”‚      Live LLM deployment + benchmarking across Gaudi3 Β· MI300X Β· NVIDIA             β”‚
β”‚      Covered: Transformer internals Β· attention Β· positional encodings Β· scaling     β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  πŸ–₯️  HPC Intern β€” CDAC Pune (National Supercomputing Mission)    Apr–Jun 2025        β”‚
β”‚      Project: High-Performance Data Management for Telecom Traffic (Berkeley DB)     β”‚
β”‚      C/C++ graph structures for networks with millions of nodes & edges              β”‚
β”‚      OpenMP parallelization Β· HPC cluster profiling Β· scalable visualization         β”‚
β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€
β”‚  πŸ§‘β€πŸ’»  Core Member & Platform Manager β€” AIHUB VVIT                       2023–Ongoing  β”‚
β”‚      Mentoring 100+ peers in applied ML Β· 3–4 hackathons/year                       β”‚
β”‚      Architected & maintain official AIHUB VVIT website & backend infra              β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

πŸ“š More Projects

Project Description Stack
🎯 SmartJob V2 Fresher jobs & internship platform β€” real-time scraping (LinkedIn, Naukri, Internshala), resume builder, intelligent caching React, TypeScript, Vite, Tailwind
🌾 FarmConnect Cross-platform agricultural mobile app with light/dark theming and advanced routing Flutter, Dart
πŸ’° Crop Price Predictor Deep learning price forecasting with RΒ² & MAPE evaluation Keras, TensorFlow, Python
πŸ“‘ Network Traffic Prediction Dual ARIMA + LSTM time-series forecasting with auto parameter selection TensorFlow, Python
πŸ’³ Amazon ML Challenge Optimal pricing for 75K+ products via multi-modal ML (catalog text + images) Python
πŸ•·οΈ Web Scraping Multi-library scraping with BeautifulSoup, Scrapy & Selenium Python
πŸ” Information Retrieval Advanced IR algorithms and search systems Python, NLP
πŸ€– Machine Learning Collection of ML algorithms and implementations Python, Scikit-learn
πŸ“Š Data Analysis Comprehensive data analysis projects Python, Pandas, NumPy

πŸ“Š GitHub Analytics

GitHub Stats GitHub Streak
Top Languages Trophies
Contribution Graph

πŸŽ“ Certifications

Certification Issuer Year
πŸ€– Generative AI (LLMs, prompt engineering, responsible AI) Google Cloud Skill Boost (L4G) 2024
β˜• Programming in Java (OOP, multithreading, exception handling) NPTEL 2024
πŸ‘₯ ACM Student Member (seminars, coding events, technical talks) ACM Ongoing

πŸ“ž Let's Connect

Open to discussing LLM systems, multi-agent architectures, HPC/GPU training, AI safety, and production AI infrastructure.

πŸ“§ ehvrreddy456@gmail.com Β |Β  πŸ“ Andhra Pradesh, India Β |Β  🌐 socialgpt.aihub-vvitu.social Β |Β  ⚑ Replies within 24 hrs


Β  Β 

Popular repositories Loading

  1. backend-practice backend-practice Public

    Python 1

  2. Web-Scraping-Project Web-Scraping-Project Public

    This project is designed to extract and process data from websites using web scraping techniques. It utilizes Python libraries such as BeautifulSoup, Scrapy, and Selenium to collect structured info…

    Jupyter Notebook

  3. aihub-vvit.github.io aihub-vvit.github.io Public

    Forked from AIHUB-VVIT/aihub-vvit.github.io

    HTML

  4. harsha4261 harsha4261 Public

    Config files for my GitHub profile.

  5. Data-Analysis-Using-Python-2024-25 Data-Analysis-Using-Python-2024-25 Public

    Jupyter Notebook

  6. SmartJob-V2 SmartJob-V2 Public

    TypeScript