Bachelors in Artificial Intelligence & Machine Learning
I build high-performance data pipelines, predictive models, and agentic AI systems. I specialize in moving beyond tutorials to architect production-grade, local-first applications that solve real engineering constraints—from processing 200M+ event streams on constrained hardware to building multi-track retrieval systems that eliminate LLM hallucinations.
- 🔭 Currently Building: Enterprise-scale predictive analytics & Agentic RAG architectures
- 💡 Core Focus: Multi-track RAG, Knowledge Graphs, Out-of-core Data Processing, NLP
- 🛠️ Recent Obsessions: Polars, DuckDB, LangGraph, KuzuDB, LLM Evaluation (RAGAS)
- 🎓 Education: B.Tech AIML @ Madhav Institute of Technology & Science (SGPA: 8.72)
🛡️ AegisLogic (PCRA) | Agentic AI & Cybersecurity
A local-first, multi-track RAG engine for Cyber Threat Intelligence. Dynamically routes queries across a Knowledge Graph (KuzuDB), Semantic Vector Store (Qdrant), and Temporal Incident Timeline (SQLite) using a LangGraph state machine and parallel asyncio fan-out. Achieved 1.0 Faithfulness on RAGAS evaluation while diagnosing a core library metric flaw.
Tech: Python, LangGraph, KuzuDB, Qdrant, Ollama, Llama 3, RAGAS, Docker
🕷️ AraneAI | Big Data & Predictive Analytics
An out-of-core analytics pipeline processing 232M+ eCommerce events (30GB+). Achieved 12× faster ingestion and 80% lower memory usage vs. Pandas using Polars lazy evaluation and Apache Parquet. Features a 20+ feature RFM store built on DuckDB to train an XGBoost churn model (AUC: 0.90), served via an async FastAPI backend with NL-to-SQL capabilities.
Tech: Python, Polars, DuckDB, XGBoost, FastAPI, Streamlit, Docker Compose
Languages:
AI / LLMs & Evaluation:
ML Frameworks:
Databases & Data Engineering:
DevOps, Tools & Soft Skills:
Let's build something scalable. Feel free to reach out via LinkedIn.