Skip to content
View Yougal17's full-sized avatar
1 follower · 2 following

Block or report Yougal17

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
Yougal17/README.md

Hey, I'm Yougal Attri 👋

AI / ML Engineer — Generative AI & Computer Vision

"I don't just build models — I build systems that work in the real world."

Portfolio LinkedIn Email

AI/ML engineer with 2+ years shipping production AI systems. I fine-tune transformer models on satellite imagery (SAM, SegFormer) and build RAG pipelines that answer questions over large document corpora — with citations, not hallucinations.

📍 Gurgaon, India · 🎓 B.Tech (Hons.) CS, Data Science — Akal University · 📄 Published with CRC Press

🔦 Featured — RBI Circular Intelligence System

Hybrid RAG pipeline over 317 RBI regulatory circulars (2022–2024, ~2.75M words) Live demo · Repo · API docs
0.96 0.91 4,368 4–9s
Answer relevancy Context precision Indexed chunks End-to-end latency
  • Scraping — custom Playwright scraper handling ASP.NET token filtering and Cloudflare-protected PDFs
  • Indexing — hierarchical chunking (120-word child / 600-word parent) into a self-hosted on-prem Qdrant database
  • Retrieval — 4 layers: dense vector search → BM25 over 9,747 terms → Reciprocal Rank Fusion → cross-encoder re-ranking
  • Generation — Gemini 2.5 Flash with enforced source citations and hallucination guarding
  • Evaluation — RAGAS across 30 questions in 12 regulatory categories

Python · Qdrant · BM25 · Playwright · Gemini 2.5 · RAGAS

🛠️ Stack

Generative AI LLMs RAG Qdrant Hugging Face Gemini

ML & Computer Vision Python PyTorch TensorFlow scikit-learn OpenCV Pandas

Deployment FastAPI Docker Google Cloud Git

Full Stack React Node.js Express.js MongoDB JavaScript SQL

💼 Experience

When Role Where
Aug 2026 – Present Executive, Network Operations Spectra, Gurgaon
Jul 2025 – Jul 2026 Fault Repair Service Engineer Nextel Communications, Gurgaon
Jan – Apr 2025 Full Stack Developer Intern (MERN) Panacea Infosec, New Delhi
Jun – Dec 2024 AI / ML Intern Panacea Infosec, New Delhi
Jan – Jun 2024 Data Science Intern Sabudh Foundation, Mohali

Highlights — fine-tuned SAM and SegFormer on geospatial satellite imagery for ~70% improvement over baseline · engineered production PyTorch/TensorFlow/FastAPI training and inference pipelines · built a 95% accuracy classification pipeline and cut analysis turnaround ~30%

Full details on the portfolio →

🚀 Other Projects

Project What it does Stack
Lead Response Assistant Multi-step LLM workflow — structured extraction, deterministic gap detection, rule-based safety guard Llama 3.1 Ollama Python
Helmet Detection · SSD Two-class SSD detector for motorcycle helmet compliance across a 764-image annotated dataset Caffe OpenCV SSD
Brain Haemorrhage Detection Hounsfield Unit–based haemorrhage detection and staging from CT DICOM images pydicom scikit-learn NumPy

📄 Publication

"Blockchain in Supply Chain Management Industry: A New Era in Manufacturing Industry" Book chapter in Intelligent Manufacturing: Exploring AI, Blockchain, and Smart Technologies in Industry 4.0 CRC Press (Taylor & Francis) · March 2025

📊 GitHub

📬 Let's Connect

Open to AI/ML engineering roles — generative AI, computer vision, and production ML.

📧 yougalattri17@gmail.com · 💼 LinkedIn · 🌐 Portfolio · 📱 +91 78074 03195

Pinned Loading

  1. rbi-rag rbi-rag Public

    Production RAG system answering questions over 317 RBI regulatory circulars with cited, verifiable sources. Hybrid retrieval + cross-encoder re-ranking.

    HTML

  2. Helmet-Detection-Using-Caffe Helmet-Detection-Using-Caffe Public

    SSD object detector on Caffe for motorcycle helmet compliance — two-class detection (with/without helmet) across a 764-image annotated dataset.

    Python

  3. Implementation-of-A-Hounsfield-value-based-approach-for-automatic-recognition-of-brain-haemorrhage- Implementation-of-A-Hounsfield-value-based-approach-for-automatic-recognition-of-brain-haemorrhage- Public

    Implementation of a Hounsfield Unit–based method for automatic brain haemorrhage detection from CT DICOM images, using a KNN classifier.

    Jupyter Notebook

  4. Lead-Response-Assistant Lead-Response-Assistant Public

    Multi-step LLM workflow for customer property enquiries — structured extraction, deterministic gap detection, and a rule-based safety guardrail layer.

    Python