AI / ML Engineer
Hyderabad
Job No. atci-5118198-s1894958
Full-time
工作描述
Project Role : AI / ML Engineer
Project Role Description : Develops applications and systems that utilize AI tools, Cloud AI services, with proper cloud or on-prem application pipeline with production ready quality. Be able to apply GenAI models as part of the solution. Could also include but not limited to deep learning, neural networks, chatbots, image processing.
Must have skills : Machine Learning (ML)
Good to have skills : Microsoft Azure Machine Learning
Minimum 12 year(s) of experience is required
Educational Qualification : 15 years full time education
Job Description: Lead ML Engineer
Role Summary
Lead the design and delivery of AI solutions across Agentic AI, Generative AI (LLMs) and classical ML/CV. Own the technical direction for suggestion & rules frameworks, search/retrieval, document and web data extraction, and image/OCR pipelines for the Value Stream. Provide architectural leadership, mentor engineers, and ensure production-grade quality, safety, and reliability. Should be familiar with evaluation strategies, responsible AI, explainability.
Responsibilities
Define end-to-end architecture for LLM/agent systems (tool use, orchestration, guardrails) and classical ML components.
Design suggestion engines and policy/rule layers that combine deterministic constraints with generative outputs.
Architect search & retrieval (BM25 + embeddings) and RAG pipelines drive relevance tuning and evaluation.
Oversee robust scraping & extraction (Playwright/Selenium/Trafilatura) and structured normalization (JSON/Parquet, schema validation).
Direct image processing and OCR workflows (OpenCV, pytesseract/ocrmypdf) for document understanding.
Establish evaluation strategy: offline/online experiments, quality/latency/cost KPIs integrate DeepEval for unit-style LLM tests.
Guide data governance, privacy/PII handling, and secure model/agent operations with MLOps partners.
Mentor the team, run design reviews, and produce clear design docs, RFCs, and POVs for stakeholders.
Concepts & Technical Awareness (Expected)
Model generalization vs. overfitting/underfitting bias/variance trade-offs regularization and early stopping.
Deep learning fundamentals: CNNs, RNNs/LSTMs/GRUs, and modern transformers encoder/decoder architectures and attention.
LLM inner-workings at a practical level: tokenization, context windows, inference strategies (batching, caching, quantization), fine-tuning/PEFT, and RAG.
Inference and serving techniques for throughput/cost (vectorization, mixed precision, compile/acceleration paths where applicable).
Tooling Familiarity
PyTorch Hugging Face ecosystem (transformers, datasets, sentence-transformers/SBERT) BERT/Llama families as applicable.
LangChain for orchestration familiarity with LangGraph/LangMem for agentic workflows (subject to approval).
spaCy, scikit-learn LightGBM/Flair where relevant Optuna for HPO SHAP for model explainability.
Search: Elastic/OpenSearch vector stores (FAISS/Pinecone/pgvector) docarray for embedding flows.
Document & web data: Playwright/Selenium, Trafilatura, pypdf, pdfplumber, pdfkit tokenization tools like tiktoken.
Stakeholder demos: Streamlit (local-only).
Qualifications
Proven record architecting and shipping production ML/LLM systems.
Strong written and verbal communication experience leading Agile delivery and cross-functional collaboration.
You will be working with a Trusted Tax Technology Leader, committed to delivering reliable and innovative solutions
Project Role Description : Develops applications and systems that utilize AI tools, Cloud AI services, with proper cloud or on-prem application pipeline with production ready quality. Be able to apply GenAI models as part of the solution. Could also include but not limited to deep learning, neural networks, chatbots, image processing.
Must have skills : Machine Learning (ML)
Good to have skills : Microsoft Azure Machine Learning
Minimum 12 year(s) of experience is required
Educational Qualification : 15 years full time education
Job Description: Lead ML Engineer
Role Summary
Lead the design and delivery of AI solutions across Agentic AI, Generative AI (LLMs) and classical ML/CV. Own the technical direction for suggestion & rules frameworks, search/retrieval, document and web data extraction, and image/OCR pipelines for the Value Stream. Provide architectural leadership, mentor engineers, and ensure production-grade quality, safety, and reliability. Should be familiar with evaluation strategies, responsible AI, explainability.
Responsibilities
Define end-to-end architecture for LLM/agent systems (tool use, orchestration, guardrails) and classical ML components.
Design suggestion engines and policy/rule layers that combine deterministic constraints with generative outputs.
Architect search & retrieval (BM25 + embeddings) and RAG pipelines drive relevance tuning and evaluation.
Oversee robust scraping & extraction (Playwright/Selenium/Trafilatura) and structured normalization (JSON/Parquet, schema validation).
Direct image processing and OCR workflows (OpenCV, pytesseract/ocrmypdf) for document understanding.
Establish evaluation strategy: offline/online experiments, quality/latency/cost KPIs integrate DeepEval for unit-style LLM tests.
Guide data governance, privacy/PII handling, and secure model/agent operations with MLOps partners.
Mentor the team, run design reviews, and produce clear design docs, RFCs, and POVs for stakeholders.
Concepts & Technical Awareness (Expected)
Model generalization vs. overfitting/underfitting bias/variance trade-offs regularization and early stopping.
Deep learning fundamentals: CNNs, RNNs/LSTMs/GRUs, and modern transformers encoder/decoder architectures and attention.
LLM inner-workings at a practical level: tokenization, context windows, inference strategies (batching, caching, quantization), fine-tuning/PEFT, and RAG.
Inference and serving techniques for throughput/cost (vectorization, mixed precision, compile/acceleration paths where applicable).
Tooling Familiarity
PyTorch Hugging Face ecosystem (transformers, datasets, sentence-transformers/SBERT) BERT/Llama families as applicable.
LangChain for orchestration familiarity with LangGraph/LangMem for agentic workflows (subject to approval).
spaCy, scikit-learn LightGBM/Flair where relevant Optuna for HPO SHAP for model explainability.
Search: Elastic/OpenSearch vector stores (FAISS/Pinecone/pgvector) docarray for embedding flows.
Document & web data: Playwright/Selenium, Trafilatura, pypdf, pdfplumber, pdfkit tokenization tools like tiktoken.
Stakeholder demos: Streamlit (local-only).
Qualifications
Proven record architecting and shipping production ML/LLM systems.
Strong written and verbal communication experience leading Agile delivery and cross-functional collaboration.
You will be working with a Trusted Tax Technology Leader, committed to delivering reliable and innovative solutions
职位要求
15 years full time education
