Skip to content
View NilayRaut's full-sized avatar
1 follower · 3 following

Achievements

Achievement: Pull Sharkx2Achievement: YOLO

Achievements

Achievement: Pull Sharkx2Achievement: YOLO

Highlights

  • Pro

Block or report NilayRaut

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.

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
NilayRaut/README.md

Nilay Raut | LLM & AI Product Engineer

Northeastern University | MS in Information Systems (Spring 2026)
Specialization: Preference Optimization (DPO), RAG Systems, & High-Performance Parallel ML

🔬 Research & Engineering Highlights

Project Key Result Stack
Aegis LLM Gateway Causal hallucination detection without ground truth; 75% cost reduction via tiered routing FastAPI, LangGraph, Ollama, DoWhy
RentSentry Rental scam detector with weighted trust scoring (LLM 60% + price signal 40%); real-time red-flag extraction; built end-to-end at SharkHack (24h) FastAPI, Claude Haiku, OpenAI, httpx, BeautifulSoup
Instruction Backtranslation 95% manual labeling reduction; 0.0622% trainable params via LoRA on A100 LLaMA-2-7B, PEFT, bitsandbytes
DPO Preference Optimization +8.8% vocabulary diversity; iterative self-rewarding loop over 2 iterations TRL, LLaMA-3.2-1B, GPT-4 API, PairRM
Parallel Video Recognition 70% E2E speedup; 15.5× CPU preprocessing via Dask on 220K-video dataset SLURM, DDP, FSDP2, I3D
RAG Financial Assistant 90% answer accuracy; +40% retrieval relevance via optimized chunking LangChain, FAISS, GPT-3.5, ada-002
FinBERT Sentiment 97.4% accuracy, 7.5× over baseline DistilBERT, HuggingFace, Gradio

🛠 Technical Stack

  • AI/ML: PyTorch, Hugging Face (Transformers, TRL, PEFT), LoRA, DPO, FAISS, LangChain.
  • Data & Performance: Dask, Joblib, NumPy, Pandas, Scikit-learn, SLURM.
  • Backend & Infra: Python (FastAPI), Redis, Docker, Fedora Linux, SQL, REST APIs.
  • Frontend: React, Next.js, Node.js.

📂 Research & Artifacts

  • Publications: Co-author of "E-Commerce Product Price Tracker using Dynamic Pricing Algorithm" (IJRASET, 2023).
  • Hugging Face: Published 4 models and 2 datasets focused on preference optimization and instruction tuning.
  • Honors: 2× recipient of the Academic Excellence Medal for top academic distinction.

📬 Connect with Me

Pinned Loading

  1. aegis-llm-gateway aegis-llm-gateway Public

    Aegis is an end-to-end production system that automatically routes prompts to the optimal LLM based on complexity, while detecting and flagging potential hallucinations using causal inference techn…

    Jupyter Notebook

  2. yashdeep94/parallel_video_recognition yashdeep94/parallel_video_recognition Public

    High-performance parallel computing for video action recognition on Something-Something V2 dataset (220k videos, 174 classes). Implementing and comparing CPU parallelization with Joblib and Dask, G…

    Jupyter Notebook

  3. finbert-sentiment-optimization finbert-sentiment-optimization Public

    This repository contains code for fine-tuning a BERT-based model for financial sentiment analysis. The project uses the Financial PhraseBank dataset to train a model that can classify financial tex…

    Jupyter Notebook 1

  4. Self-Alignment-with-Instruction-Backtranslation Self-Alignment-with-Instruction-Backtranslation Public

    LLaMA-2-7B self-alignment via instruction backtranslation. 95% labeling reduction, 0.0622% trainable params (LoRA r=8), 3 HuggingFace artifacts published.

    Jupyter Notebook

  5. rentsentry rentsentry Public

    AI-powered rental listing fraud detector — paste a Craigslist URL or listing text, get a trust score, red flags, and neighborhood context

    HTML