Data Scientist/Machine Learning Engineer

Description :

Role : Data Scientist / Machine Learning Engineer (58 Years).

Location : Chennai.

Time Zone : IST.

Primary Tech Stack : AI/ML, Algorithms, Deep Learning, Python, LLMs, CI/CD, APIs.

Experience Required :

- Must-Have : 5+Years of hands-on experience in ML/Deep Learning, including 3+ years in a senior or lead role.

Practical experience with OCR pipelines :

- Tesseract, EasyOCR, PaddleOCR, AWS Textract, Google Vision.

- Experience with Document Layout Models :

- LayoutLMv3, Donut, Pix2Struct, LiLT, LayoutXLM.

Expertise in Vision-Language Models :

- CLIP, BLIP-2, GroundingDINO, LLaVA, Florence-2.

Strong background in LLM-based extraction using :

- OpenAI, Gemini, Claude, Llama, Qwen, Mistral.

Proven ability to design end-to-end ML system architectures, including :

- LLM + OCR + Embedding + Prompt pipelines.

- Preprocessing pipelines for multi-format documents.

- Vector DBs, embedding stores, structured extraction systems.

- Async processing queues, job orchestration, microservices.

- GPU/CPU deployment strategy.

Experience scaling ML systems :

- Batch processing large files.

- Handling concurrency, throughput, latency.

- Model selection, distillation & quantization (GGUF, ONNX).

Preferred Experience :

- CI/CD for ML : GitHub Actions, Jenkins.

- Model monitoring : latency, drift detection, cost optimization.

- Cloud experience (AWS/GCP/Azure), including : SageMaker, Vertex AI, Bedrock (nice to have).

- Ability to choose the right ML approach : fine-tuning, RAG, multimodal, prompting.

- Strong problem-solving and end-to-end solution ownership.

- Skilled in building PoCs and converting them into production-ready systems.

- Ability to estimate feasibility, cost, complexity & timelines for ML solutions.

Education :

- Required : Bachelors or Masters Degree in a relevant field.

Key Responsibility Areas (KRAs) :

- Design and implement scalable ML pipelines integrating OCR + Document Layout Models + LLMs + Embeddings.

- Architect high-performance processing pipelines for multi-format documents (PDF, scanned docs, images, compound files).

- Own system design decisions including model selection, hardware strategy, vector DB usage, orchestration flow, and API layer.

- Build, enhance, and optimize OCR pipelines using Tesseract, EasyOCR, PaddleOCR, AWS Textract, Google Vision.

- Develop and fine-tune LayoutLMv3, Donut, Pix2Struct, LiLT, LayoutXLM for structured data extraction and document understanding.

- Implement and integrate Vision-Language Models (CLIP, BLIP-2, GroundingDINO, LLaVA, Florence-2) for multimodal extraction.

- Build robust LLM extraction pipelines using OpenAI, Gemini, Claude, Llama, Qwen, Mistral.

- Own prompt engineering, schema-driven extraction, evaluation frameworks, and hybrid pipelines (RAG + multimodal + OCR).

- Optimize LLM performance through distillation, quantization (ONNX, GGUF), and fine-tuning where needed.

Architect pipelines for high-volume document processing with focus on :

- Throughput

- Latency

- Resiliency

- Concurrency management

- Build async queues, microservices, and load-balanced processing flows.

- Optimize deployments across CPU/GPU environments (containerized or serverless).

- Set up automated model deployment pipelines using GitHub Actions, Jenkins, etc.

Implement monitoring dashboards for :

- Model drift

- Latency

- Accuracy

- Cost and resource optimization

- Establish feedback loops for model re-training and continuous improvement.

- Work closely with product, engineering, QA, and DevOps teams to deliver production-grade ML solutions.

- Translate business/functional requirements into ML problem statements and technical designs.

- Present architecture, feasibility, estimates, and timelines to leadership and stakeholders.

- Quickly build Proof-of-Concepts and evolve them into full-scale, production-ready systems.

- Evaluate multiple ML approaches (RAG, multimodal, prompting, fine-tuning) and recommend the most effective solution.

- Ensure code quality, documentation, reproducibility, and scalable design.