Skip to content
View Sahilkumar-r's full-sized avatar

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull SharkAchievement: YOLO

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull SharkAchievement: YOLO

Block or report Sahilkumar-r

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.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. 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
Sahilkumar-r/README.md

About me

Recent Data Science graduate (BS, FAST NUCES, 2026, CGPA 3.84) who builds retrieval, agent and model projects and measures them. I report what the numbers say, including the results that did not go my way.

Featured projects

FastAPI Docs Assistant

Status RAG Python FastAPI

Answers questions about the FastAPI docs, cites the pages it used and refuses off topic questions.

  • Correct page in the top five for 30 of 30 test questions (hit@5 1.00, MRR 0.89 vs 0.77 for keyword search alone)
  • Refused 8 of 8 off topic questions, wrongly refused 0 of 30
  • Hybrid search runs in about 9 ms on CPU

View code

Multimodal RAG agent

Status Agents LangGraph Guardrails

LangGraph agent with Corrective RAG, guardrails and tracing across three model providers.

View code

GPT-2 distilled from Mistral 7B

Status Distillation PyTorch

Cross tokenizer knowledge distillation with an honest write up of a negative result.

  • 99% of token positions aligned across tokenizers
  • Top 1 agreement 36.7% vs 33.6% baseline
  • Perplexity 28.98 to 29.64: no improvement, reported as is

View code

Machine learning from scratch

Status NumPy scikit-learn

Fifteen NumPy implementations, benchmarked against scikit-learn.

  • Logistic regression 97.5% vs 98.1%, KNN 97.4% vs 97.4%
  • Linear regression and PCA match scikit-learn exactly
  • Trees trail on small data and are far slower, and the table says so

View code

Also built: LangGraph chatbot with a confidence guardrail · Vision Transformer service · Bagging studio · Voting studio · MCP multi server agent

In progress: next word LSTM and a gated RNN from scratch. Results will be added once I have run them.

More on the portfolio, with the problem, what I built and the result for every project.

Skills

Languages and ML

Python, C++, Bash, PyTorch, scikit-learn, NumPy, Pandas, R

LLMs, APIs and data

FastAPI, PostgreSQL, MySQL, MongoDB, Redis, GraphQL, Hugging Face, LangChain

Cloud and DevOps

Docker, Kubernetes, AWS, Azure, GitHub Actions, Linux, Git, GitHub

LLMs RAG Agents Vector DBs Data MLOps

Let's talk

I am looking for AI Engineer and Forward Deployed Engineer roles where I can build with customers and ship measured results.

Email me Portfolio

Popular repositories Loading

  1. First-Production-RAG First-Production-RAG Public

    A production-ready Retrieval-Augmented Generation (RAG) pipeline built with FastAPI, Inngest, Qdrant Cloud, and Groq, optimized for free-tier deployment.

    Python

  2. Langchain-Basics Langchain-Basics Public

    I have started learning langchain from scratch, So will be building something amazing soon

    Jupyter Notebook

  3. Simple-RAG Simple-RAG Public

    Used the rag locally to build the basics like creating Vector DB, Retrieval, and managing History.

    Python

  4. Machine-Learning-From-Scratch Machine-Learning-From-Scratch Public

    For the next course of days I will be building ML models From scratch.

    Python

  5. bagging-ensemble-studio bagging-ensemble-studio Public

    Streamlit app to build and evaluate bagging ensembles on your own CSV, for classification and regression

    Python

  6. voting-ensemble-studio voting-ensemble-studio Public

    Streamlit app to build and evaluate hard and soft voting ensembles on your own CSV

    Python