H
Haider Ali Shah
About
Detail
California, United States
Résumé
Jobs
verified_user
0% verified
Lead AI/ML ArchitectPineconeJan 2022 - Current (4 years 10 months)Architected enterprise-scale AI/ML solutions including NLP, RAG, and GenAI applications, defining microservices and serverless architectures. Led AI/ML strategy, system design, and roadmap for scalable cloud-native deployments using AWS, GCP, and Azure. Established end-to-end MLOps pipelines with MLflow, W&B, Evidently AI, and Prometheus/Grafana for monitoring, experiment tracking, and model governance. Oversaw deployment of transformer-based LLM systems for summarization, question answering, and enterprise chatbots. Implemented CI/CD pipelines and automated validation for model packaging, drift detection, and retraining workflows. Applied advanced optimization techniques (operator fusion, TorchDynamo, TorchInductor) to improve LLM inferenc- CSenior Machine Learning EngineerCodeNinja Inc.Jul 2018 - Dec 2021 (3 years 6 months)Led development of enterprise-scale Generative AI solutions leveraging transformer-based LLMs (BERT, GPT, T5) for summarization, semantic search, and conversational AI. Built and optimized RAG pipelines integrated with vector databases (FAISS, Pinecone) for real-time knowledge retrieval. Applied LORA, PEFT, and parameter-efficient fine-tuning for large transformer models on limited compute resources. Developed production-grade ML APIs using FastAPI, containerized with Docker and Kubernetes for scalable deployment. Integrated LangChain to create modular LLM-powered applications with structured output parsing, tool usage, and memory augmentation. Conducted kernel-level profiling using Kineto trace and Torch dispatcher to optimize GPU executio
- Machine Learning EngineerCognitiveScaleAug 2015 - Jun 2018 (2 years 11 months)Built time-series forecasting and anomaly detection models using LSTM, GRU, ARIMA, and statistical techniques. Integrated pre-trained models (BERT, XLNet) for multilingual NLP applications including text classification, summarization, and entity extraction. Developed modular ML pipelines with feature engineering, hyperparameter tuning, and automated retraining using Airflow, MLflow, and Docker. Deployed models in production environments using Flask APIs, TensorFlow Serving, and containerized microservices. Implemented active learning and semi-supervised techniques to improve labeled dataset quality and reduce annotation effort. Designed experiments and evaluation metrics to measure model robustness, generalization, and performance drift. Co
- Data ScientistHortonworksSep 2013 - Jul 2015 (1 year 11 months)Developed predictive models for customer churn, product recommendation, and operational KPIs using Python, R, Scikit-learn, and XGBoost. Built NLP pipelines for sentiment analysis, text classification, and named entity recognition using NLTK, spaCy, and TextBlob. Engineered ETL pipelines and automated feature extraction for structured and unstructured datasets. Deployed models through REST APIs using Flask, enabling real-time prediction for downstream applications. Designed dashboards and visualizations using Matplotlib, Seaborn, and Tableau to present insights to stakeholders. Conducted A/B testing and model validation to measure performance using metrics such as accuracy, F1, AUC, and RMSE. Collaborated with cross-functional teams to tran
Education
verified_user
0% verified
- BBachelor of Computer Science
Projects (professional or personal)
verified_user
0% verified
- GGenerative AI Summarization EngineBuilt a large-scale document summarization pipeline using LLMs (GPT, T5) and LangChain for modular GenAI applications in enterprise settings. Deployed the system on cloud GPU clusters using FastAPI and Kubernetes, enabling real-time API access and improving summarization efficiency by 40%.
- EEnterprise Conversational AI PlatformBuilt an end-to-end LLM-powered chatbot using LangChain and RAG, capable of structured output parsing, tool usage, and memory augmentation. Deployed on cloud GPU-backed clusters with FastAPI and Kubernetes, improving customer query resolution by 55%.
- RRAG-Based Knowledge Retrieval SystemDeveloped a real-time document search pipeline integrating LLMs (GPT, T5) with vector databases (FAISS, Pinecone) to provide accurate contextual responses. Optimized inference latency by 35% using GPU profiling, kernel-level optimizations, and parameter-efficient fine-tuning techniques.