Data Scientist & AI Systems Engineer
Agentic AI · Statistical Machine Learning · NLP · Production Data Pipelines

I build machine learning and LLM systems that make it all the way to a deployed, user-facing product — not just a notebook.

My work sits at the intersection of three things: autonomous agent workflows (LangChain / LangGraph), statistically defensible modelling (hypothesis testing, feature validation, knowing when OLS beats GBM), and getting messy real-world data through web scraping and APIs before any of the modelling starts.

• 🎓 BCA in Data Science, SRM University — CGPA 9.5/10
• 🧪 Delivered an NLP fake-news classifier at 94%+ accuracy
• 🤖 Built and deployed a 4-agent autonomous research system on Streamlit Cloud
• 📊 Data Analyst Intern — automated 10+ KPI dashboards, cutting manual reporting effort ~35%
• 💬 Open to roles in AI Engineering, Data Science, and Machine Learning

LangChain · LangGraph · Groq LLM · Tavily · BeautifulSoup · Streamlit

Four cooperating agents — search → reader → writer → critic — that turn a single topic into a fact-checked research report with no human in the loop. The search agent finds sources, the reader scrapes and cleans them, the writer drafts a structured report, and the critic reviews it for gaps before output.

Runs as an interactive Streamlit app or a CLI pipeline, with live output streaming. 🔗 Live demo

DeepFace · OpenCV · CNN · YouTube Data API v3 · Streamlit

Detects a user's dominant facial emotion from an uploaded photo and returns a personalised YouTube playlist in under 5 seconds. Seven emotion classes, each mapped to a curated music query. Taken from Jupyter prototype to a live, publicly accessible web app.

SQL (SQLite) · Python · Power BI · DAX · Excel

End-to-end analysis of a 9,994-row retail dataset. Advanced SQL window functions surfaced that 18.7% of orders were loss-making and that discounts above 40% consistently produced negative profit. Delivered as a multi-page Power BI dashboard with cross-filtering slicers and drill-downs.

Python · Scikit-learn · Pandas · Statistical Testing

Predictive pipeline where every feature had to earn its place: Pearson correlation and Chi-Square testing validated feature selection, with p-values as the quantitative basis for keep/drop decisions. Identified smoking status as the strongest cost driver, separating genuine predictive signal from correlation.

Languages   Python SQL R

AI & Agents   LangChain LangGraph Groq LLM RAG Architecture Tool Calling Streamlit

Machine Learning   Scikit-learn Linear & Logistic Regression Decision Trees Gradient Boosting Model Evaluation

NLP & Vision   Text Classification TF-IDF DeepFace OpenCV CNNs

Statistics   Hypothesis Testing Chi-Square Pearson Correlation Correlation vs. Causation EDA

Data Engineering   Pandas NumPy Web Scraping (BeautifulSoup, Tavily) Feature Engineering Data Cleaning

BI & Visualisation   Power BI Tableau DAX Matplotlib Seaborn Excel

Databases & Tools   MySQL SQLite MongoDB Git GitHub

Data Analyst Intern — Early-stage B2B Startup · Oct 2025 – Dec 2025 Sales & Engagement Analytics

• Cleaned and ran EDA on 3 real-world business datasets, surfacing 4 user-behaviour patterns the product team flagged for roadmap decisions
• Built Power BI and Tableau dashboards tracking 10+ KPIs, reducing manual reporting effort by ~35% via automated pipelines
• Presented findings to non-technical stakeholders, leading to 2 adopted operational process changes

Open to opportunities in AI Engineering, Data Science, and Machine Learning.
📫 himanshirathore25102005@gmail.com