👋 Hey there! I'm Ansar Afsar, a Master’s in Computer Science grad (Government Arts College, Salem, 2025, 8.44 GPA) driven by a passion for building AI-driven, privacy-first solutions. From multilingual NLP pipelines to scalable cloud deployments, I craft tools that solve real-world problems with precision and impact.
My mantra? Turn complex data into actionable insights using Python, PyTorch, and a sprinkle of DevOps magic.

• 🔭 Current Focus: Actively seeking roles in developing AI/ML pipelines for content moderation and personalization, emphasizing privacy and scalability. I’m passionate about rapid prototyping—building and breaking things quickly, creating 10 projects a day to satisfy my hunger for growth.
• 🌱 Learning Edge: Deepening expertise in NLP (transformers, embeddings), MLOps, and AWS cloud orchestration.
• 🎓 Education:
  • M.S. Computer Science (2025)
  • B.S. Computer Science (2023)
  • Government Arts College, Salem
• 🏆 Certifications:
  • Google Data Analytics
  • IBM Data Analyst
  • Cisco Networking
  • Microsoft Business Analytics

https://ansarafsar.github.io/portfolio/

[

A GPU-accelerated microservices pipeline for real-time content moderation on professional platforms.

Processes JSON inputs (e.g., "Ami kal party te jabo") with:

• Translation: easynmt (m2m_100_418M) for non-English to English
• Hate Speech Detection: Ensemble (fasttext, MuRIL, IndicBERT, XLMR, DistilBERT) with rule-based boosts; flags >0.7 scores
• Tagging: intfloat/e5-large-v2 embeddings for personalized curation

Tech: Docker, PyTorch 2.1.2, AWS EC2 g4dn.xlarge (NVIDIA T4, CUDA 12.8), Nginx proxy
Impact: Sub-0.8s inference, fault-tolerant with backup models, logs to flagged_hate_speech.jsonl for retraining
Privacy: Self-hosted to bypass external APIs like GPT, ensuring data security

A Streamlit app forecasting retail trends using Scikit-learn and Pandas.
Impact: Optimized inventory planning with predictive analytics.
Demo: [GIF TBD]

AI-driven Q&A for retail analytics using LangChain and Streamlit.
Impact: Instant insights from complex datasets for business users.

Blockchain-based logistics tracking on Algorand testnet with AlgoKit.
Impact: Enhanced transparency in supply chain workflows.

NLP-driven social media trend visualization using NLTK and Seaborn.
Impact: Empowered marketing teams with sentiment insights.

• 💼 Internships: Contributed to AI/ML projects at ZEEX AI, Tienext, BOOKDIO, and Unified Mentor (2024–2025)
• 🛠️ Open Source: Fixed bugs and improved docs for Python ML libraries
• 👨‍🏫 Mentorship: Guided 15+ undergrads in CS projects, focusing on AI and data science
• ✍️ Blogging: Published 10+ Medium articles on AI trends, NLP, and cloud computing

My work focuses on privacy-first AI, leveraging self-hosted models to avoid data leaks (unlike GPT-based APIs).
The hate speech pipeline showcases:

• Scalability: Modular microservices for easy scaling on AWS
• Efficiency: GPU-accelerated inference (<0.8s) for real-time apps
• Impact: Safer digital platforms through robust moderation and curation

• 🚴 Cycling trails keep my code sharp and spirit high
• 🎲 Strategy games are my jam—love outsmarting AI opponents
• 📚 Always reading: AI papers, tech blogs, or a thrilling sci-fi novel

⭐ Star my repos if you find them inspiring! Let’s build the future of AI together!

✍️ Portfolio Website