Skip to content
View Ayushforai's full-sized avatar
💭
🍵 Deploying from cafe
💭
🍵 Deploying from cafe

Highlights

  • Pro

Block or report Ayushforai

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. Markdown is supported. 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
Ayushforai/README.md

Hi, I'm Ayush Chaudhary 👋

AI/ML Engineer — I fine-tune and deploy LLMs, not just study them


🚀 About Me

I don't just study AI/ML — I ship it, break it, and fix it until the numbers move.

  • 🎓 Final-year CSE undergrad at Amity University, Greater Noida (2022–2026)
  • 🏎️ Building Racecoe, a hybrid Formula 1 Q&A assistant — closed 19/20 (95%) tracked correctness issues, shipped with 230+ regression tests.
  • 🗣️ Fine-tuned LLaMA 3.2 with LoRA/QLoRA — cut trainable params 1.3B → 20.9M, hit 89.5% token accuracy, deployed live on Hugging Face Spaces.
  • 🔍 I build retrieval pipelines that don't hallucinate (FAISS, multi-source RAG routing).
  • 📦 I ship to production, not notebooks — Docker, Render, CI smoke tests.
  • 💼 Open to AI/ML Engineering roles — internship or full-time.
  • 📫 Reach me at ayushc90210@gmail.com

🛠️ Tech Stack

Python PyTorch HuggingFace FastAPI React Docker MySQL Git


📌 Featured Projects

🏎️ Racecoe — Hybrid F1 Chatbot · Aug 2026 – Present
Hybrid F1 Q&A assistant (FastAPI + React) routing across the OpenF1 API, historical race CSVs, and a FAISS regulation/race RAG index, with citations and multi-turn follow-ups. Closed 19/20 (95%) tracked correctness issues; shipped with 230+ regression tests and deploy smokes (in-process, Docker, live Render). Runs on Gemini 3.8 Flash with retry/fallback handling.

🗣️ LLM Fine-Tuning for Slang & Informal Text Translation · Jul 2025 – Apr 2026
Fine-tuned LLaMA 3.2 1B Instruct with QLoRA (rank 16) — 11.27M trainable params (0.91% of 1.24B) for formal → slang rewriting. Training loss 4.75 → 0.267 (94.4% drop), 89.5% token accuracy over 2,106 steps. Built a formality detector (TF-IDF + logistic regression) hitting 91.1% accuracy/F1. Deployed as a Flask app on Hugging Face Spaces.

🎬 Movie-Sentiment-Analysis · Aug 2025 - Feb 2026
A Natural Language Processing (NLP) application designed to perform sentiment analysis on movie reviews. Using the IMDB dataset of 50,000 reviews, the system preprocesses raw text data, converts it into numerical features using TF-IDF, and trains a Logistic Regression model to classify the sentiment. The final model is deployed in a lightweight Flask web application, allowing users to input their own reviews and receive instant predictions.


📜 Certifications

  • DeepLearning.AI — Agentic AI (Feb 2026)
  • Oracle — Cloud Infrastructure 2025 Certified AI Foundations Associate (Jan 2026)
  • Simplilearn — RAG Course for Beginners (Nov 2025)

Currently exploring: Advanced system design and multi-agent RAG pipelines.

Pinned Loading

  1. Racecoe Racecoe Public

    A hybrid Formula 1 assistant (formerly the F1 Pit Wall chatbot) that routes questions to the right data source: live telemetry (OpenF1), structured historical CSV lookups, or RAG over FIA regulatio…

    Python

  2. Movie-Sentiment-Analysis Movie-Sentiment-Analysis Public

    A modern web application for analyzing movie review sentiment using machine learning with sap green and gold UI theme

    HTML

  3. Slang-Translator Slang-Translator Public

    AI-powered formal to informal text translator. A Flask app, Llama 3.2 1B + old LoRA adapter on Hugging Face,

    Python

  4. abstractive-dialogue-summarizer abstractive-dialogue-summarizer Public

    Abstractive Dialogue Summarizer using T5