Looking only USA local candidates.
Location- Remote
Key Responsibilities
• Strong Data Engineer with Agentic AI experience, capable of Data Extract, Transformation, Feature Engineering, Analytics to build AI/ML models
• Curate and preprocess training corpora for domain-specific instruction tuning.
• Fine-tune open-source LLMs using LoRA, RLHF, DPO, and model distillation techniques.
• Implement model evaluation pipelines and benchmark reporting.
• Collaborate with Prompt & Data teams to create repeatable model tuning workflows.
1. Data Engineering & Preparation
• Architect and implement data pipelines for large-scale text ingestion, cleaning, and transformation.
• Perform data extraction, transformation, and feature engineering across structured and unstructured sources.
• Develop and maintain data quality frameworks ensuring clean, diverse, and bias-mitigated datasets for model training.
• Automate data labeling and annotation workflows using LLM-assisted or agentic tools.
• Build domain-specific corpora for instruction tuning, conversational grounding, and retrieval-augmented training.
2. Model Training & Fine-Tuning
• Fine-tune and adapt open-source LLMs (e.g., LLaMA, Mistral, Falcon, Gemma) using LoRA, QLoRA, RLHF, DPO, and model distillation.
• Implement self-instruct and multi-turn conversational fine-tuning for agentic use cases.
• Design training orchestration scripts for distributed GPU/TPU environments (PyTorch, DeepSpeed, HuggingFace Accelerate).
3. Model Evaluation & Benchmarking
• Develop evaluation frameworks for automatic and human-in-the-loop assessment of LLM performance.
• Benchmark models against standard datasets (MMLU, HELM, ARC, TruthfulQA) and custom internal benchmarks.
• Generate detailed performance dashboards tracking precision, hallucination rate, factual consistency, and latency.
• Conduct A/B testing and regression analysis on model updates to ensure stable improvement.
4. Collaboration & AI Workflow Automation
• Work cross-functionally with Prompt Engineers, Data Scientists, and DevOps to operationalize model development.
• Build repeatable pipelines for fine-tuning, version control, and continuous model improvement (MLOps).
• Integrate agentic feedback loops for continuous self-improvement and autonomous retraining cycles.
• Support deployment through containerized model serving (FastAPI, Triton, or Ray Serve).
5. Research & Innovation
• Stay current with cutting-edge research in LLM fine-tuning, alignment, and model compression.
• Contribute to internal whitepapers and experiments evaluating emerging architectures and optimization methods.
• Prototype and publish novel training methodologies or agentic evaluation techniques.
Required Skills & Experience
• Strong Python expertise with hands-on experience in PyTorch, Hugging Face Transformers, and LangChain.
• Deep understanding of LLM architectures, tokenizer mechanics, and parameter-efficient fine-tuning.
• Proficiency in data processing frameworks (Spark, Airflow, Pandas, Arrow, Dask).
• Experience with distributed training and GPU/TPU optimization (CUDA, NCCL).
• Knowledge of evaluation metrics and human-aligned reward modeling.
• Experience with Vector Databases (FAISS, Milvus, Pinecone) for context retrieval.
• Familiarity with cloud platforms (AWS, GCP, Azure) and container orchestration (Docker, Kubernetes).
• Exposure to agentic AI frameworks and feedback-based continuous improvement systems is a plus.
Preferred Qualifications
• Prior experience contributing to open-source LLM projects.
• Background in NLP research or applied ML.
• Knowledge of data privacy, ethical AI, and prompt alignment techniques.
• Master’s or Ph.D. in Computer Science, AI, or related field preferred.
What You’ll Get to Build
• A home-grown, domain-specialized LLM trained on proprietary and public datasets.
• A scalable fine-tuning pipeline that powers multiple downstream agents and AI applications.
• An autonomous model training framework capable of learning from feedback in real time.