AI / ML Engineer · Gurgaon, India

Yougal
Attri

AI/ML engineer with 2+ years of experience shipping production AI systems across computer vision and generative AI. I fine-tune transformer models on satellite imagery, and build retrieval-augmented systems that answer questions over large document corpora — every answer traced back to its source.

Get in touch LinkedIn Resume
GitHub +91 78074 03195
YA

Technical Skills

🧠 Machine Learning
Deep Learning Neural Networks Transformers Fine-Tuning Transfer Learning Model Evaluation
✦ Generative AI
LLMs RAG Prompt Engineering Embeddings Hybrid Search Qdrant Chatbots
🛰️ Computer Vision
Image Segmentation SAM SegFormer Satellite Imagery OpenCV
⚙️ Pipelines & Deployment
FastAPI Docker REST APIs Google Cloud On-Prem Deploy End-to-End ML
💻 Frameworks & Languages
Python PyTorch TensorFlow Hugging Face scikit-learn Pandas NumPy SQL C/C++
🌐 Full Stack
JavaScript React.js Node.js Express.js MongoDB JWT Auth Git/GitHub

Experience

Executive — Network Operations

Spectra · Gurgaon

Aug 2026 – Present
  • Enterprise network operations — monitoring, escalation, and service continuity.

Fault Repair Service Engineer

Nextel Communications · Gurgaon

Jul 2025 – Jul 2026
  • Handled SLA-driven fault resolution for enterprise and retail customers.
  • Maintained tracking and reporting on fault trends, resolution times, and recurring failure causes for weekly operational review.

Full Stack Developer Intern (MERN)

Panacea Infosec Pvt. Ltd. · New Delhi

Jan – Apr 2025
  • Built responsive frontend modules for a cybersecurity audit web app using React.js, HTML, and CSS.
  • Implemented RESTful APIs with Node.js & Express.js; integrated JWT-based auth with role-based access control.
  • Collaborated with cross-functional teams to translate stakeholder requirements into technical solutions.

AI / ML Intern

Panacea Infosec Pvt. Ltd. · New Delhi

Jun – Dec 2024
  • Fine-tuned SAM and SegFormer transformer networks on geospatial satellite imagery, improving segmentation performance roughly 70% over baseline.
  • Applied transfer learning and loss-function tuning to handle severe class imbalance in remote sensing data.
  • Engineered production-grade Python training and inference pipelines (PyTorch, TensorFlow, FastAPI), moving models from experimentation into deployment.

Data Science Intern

Sabudh Foundation · Mohali

Jan – Jun 2024
  • Designed supervised and unsupervised ML models on social-impact datasets.
  • Built a classification pipeline reaching 95% accuracy.
  • Streamlined data preparation workflows with Pandas and scikit-learn, cutting analysis turnaround by roughly 30%.

Projects

RBI Circular Intelligence System

Hybrid RAG pipeline over 317 regulatory circulars

Flagship

An end-to-end retrieval-augmented generation system built in Python over 317 RBI circulars from 2022–2024 — roughly 2.75 million words. It covers the full chain: scraping, parsing, chunking, retrieval, and generation. Every answer cites the circular it came from, along with its date and URL.

0.96

Answer relevancy (RAGAS)

0.91

Context precision

4,368

Indexed chunks

4–9s

End-to-end latency

  • Custom Playwright scraper handling ASP.NET token-based filtering and Cloudflare-protected PDF delivery.
  • Hierarchical chunking (120-word child / 600-word parent) indexed in a self-hosted on-prem Qdrant vector database.
  • Four-layer hybrid retrieval: dense vector search, BM25 over a 9,747-term vocabulary, Reciprocal Rank Fusion, and cross-encoder re-ranking.
  • Gemini 2.5 Flash generation layer enforcing citations and hallucination guarding; tuned with RAGAS across 30 questions in 12 regulatory categories.
Python Qdrant BM25 Playwright Gemini 2.5 RAGAS
Live demo View repo

Lead Response Assistant

Multi-step LLM workflow for customer property enquiries. Extracts structured JSON first, detects missing fields deterministically in Python, then generates advice behind a rule-based safety guard.

Llama 3.1 · Ollama · Guardrails

View on GitHub

Helmet Detection · SSD on Caffe

Two-class SSD object detector for motorcycle helmet compliance, trained end-to-end from Pascal VOC annotations through LMDB conversion to annotated inference output.

764 images · Caffe · OpenCV

View on GitHub

Brain Haemorrhage Detection

Hounsfield Unit–based detection and staging of intracranial haemorrhage from CT DICOM data, with HU-window segmentation and a KNN classification stage.

DICOM · pydicom · scikit-learn

View on GitHub

Education & Publication

Education

B.Tech (Honors), Computer Science & Engineering

Data Science Specialization
Akal University, Bathinda · 2020 – 2024

CGPA 8.36 / 10

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

Contact

✦ Open to opportunities

Let's build
something great.

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

Email Me LinkedIn Call Me