Data Science · Data Analysis · Machine Learning · Deep Learning · NLP · Generative AI · Agentic AI

Fluent across the AI stack, from data analysis and machine learning to retrieval-augmented generation and autonomous agent systems.
The work spans the full path from raw data to a working system: understanding and cleaning the data, building and comparing models,
and turning them into applications with retrieval, tools and controlled agent behavior.

• Exploratory data analysis, SQL querying, data cleaning, outlier and missing-value handling
• Feature engineering, feature selection and dimensionality reduction (PCA)
• Visualization with Matplotlib and Seaborn to find structure, relationships and data-quality issues before modeling

• Supervised learning: regression and classification with Logistic Regression, KNN, Naive Bayes, Decision Trees, SVM and gradient boosting (XGBoost)
• Unsupervised learning: clustering and dimensionality reduction
• Model evaluation, hyperparameter tuning and systematic comparison against baselines
• End-to-end pipelines: cleaning, feature engineering, tuning and evaluation (Ames Housing regression with a tuned XGBoost model: RMSE 0.1308, R² 0.9083)

• Architecture benchmarking: Perceptron, ANN and CNN compared on MNIST to study how architectural choices, not only depth, drive performance
• Sequence modeling: a SimpleRNN baseline against an LSTM to show the effect of gated memory
• Neural network design, training and evaluation with TensorFlow

• Text preprocessing, tokenization and embeddings
• Text and emotion classification (six-class emotion prediction from raw sentences)
• Language modeling and next-word prediction with recurrent architectures
• Transformer-based models and the Hugging Face ecosystem

• Retrieval-augmented generation: document loading, chunking, embedding-model selection (HuggingFace embeddings including KURE-v2), vector indexing in Chroma and FAISS, retrieval, and answer generation grounded in retrieved context with Groq-hosted LLMs
• Retrieval quality: evaluating RAG pipelines on retrieval accuracy and latency, and tuning context retrieval to improve both
• Structured outputs: function calling and schema-constrained JSON output so model responses can drive deterministic downstream steps
• Prompt engineering: instruction design, output formatting and prompt iteration against evaluation results
• Speech and media pipelines: media ingestion with yt-dlp, audio chunking, local transcription with OpenAI Whisper, then LLM summarization and structured extraction of action items, key decisions and open questions
• Conversational interfaces: RAG chains over transcripts and documents, served through Streamlit
• LLM providers and SDKs: Anthropic Claude API, Groq, LangChain

• Workflow design in LangGraph: typed state schemas, nodes and edges, sequential pipelines, and parallel branches merged with reducers
• Routing: conditional edges that classify a query and direct it to the right pipeline (handbook retrieval, fee-structure retrieval or general knowledge), with deterministic routing loops
• Iterative loops: tool-use cycles with a reviewer step that evaluates output and sends it back for revision
• Tool use: tool and function-calling mechanisms with validated, structured arguments and results
• Human-in-the-loop: approval checkpoints that pause a workflow for a person to review or edit before it continues
• Multi-agent orchestration: coordinating specialized agents with explicit state management so behavior stays predictable and traceable
• Frameworks: LangGraph, LangChain, Anthropic Claude SDK

Languages and data

Machine learning and deep learning

Generative AI and agentic AI

Environments and tools

Open to opportunities across data science, data analysis, machine learning, deep learning, NLP, generative AI and agentic AI, in India, remote and internationally.

LinkedIn · Kaggle · pathanalfaiz72@gmail.com