Semil Periyasamy

AI/ML Engineer with 2 years of experience building production-grade agentic systems, end-to-end LLM pipelines, and high-throughput inference infrastructure. Skilled in prompt engineering, embeddings, and fine-tuning large language models (Llama 3 70B) with HuggingFace Transformers, orchestrating RAG pipelines with LangChain and LlamaIndex, and implementing vector search with FAISS and Milvus. Experienced in model evaluation and benchmarking, CI/CD and version-controlled deployment workflows, and responsible AI practices covering governance, security, and data privacy.

Builds backend services with FastAPI and data pipelines with PySpark, MongoDB, and SQLAlchemy; deploys Gemini-based document intelligence solutions on GCP and Azure.

Machine Learning Engineer at AK Technologies (2024-09 – Present)

• Architected and deployed a production multi-agent ERP assistant with FastAPI-based backend services, enabling autonomous, document-driven finance queries and workflow automation, following version-controlled (Git) development and CI/CD deployment practices.
• Fine-tuned Llama 3 70B and BERT-based models using HuggingFace Transformers and prompt engineering on synthetic datasets, improving inference throughput by 40% with TensorRT-LLM, vLLM, INT8/FP8 quantization, and KV-cache optimization.
• Built distributed GPU inference infrastructure with Redis caching, reducing multi-node latency by 35%, and implemented health checks and fallback mechanisms for production reliability.
• Developed a multilingual document-intelligence system on GCP/Azure using Gemini 1.5, processing 1,000+ high-resolution requests per hour with full W&B observability for latency and model-drift monitoring, and structured model evaluation for accuracy tracking.
• Led RAG pipeline development for a geospatial urban-transport planning platform, orchestrating retrieval and embeddings with LangChain/LlamaIndex and FAISS/Milvus vector stores, with supporting ETL pipelines using PySpark, MongoDB, and SQLAlchemy-backed storage, applying data validation, governance, and security checks for accuracy and consistency.
• Designed a Tamil-language Vision-OCR pipeline for structured extraction and LLM-powered translation, delivering document analytics in a regulated, audit-critical domain with attention to data privacy and responsible AI handling.

AI Engineer Intern at DigitalGarage Pvt. Ltd (2024-03 – 2024-08)

• Developed and benchmarked computer vision and NLP models using Python and TensorFlow, evaluating performance against standard benchmarks and unit-testing model outputs prior to deployment.
• Contributed to full-stack AI problem-solving initiatives, supporting system design and model evaluation pipelines.

Data Science Intern at Shiash Info Solutions Pvt. Ltd (2024-02 – 2024-05)

• Performed multi-source data analysis using Python (Pandas, NumPy) and built Seaborn/Matplotlib visualizations to communicate insights to non-technical stakeholders.

B.Tech in Artificial Intelligence & Data Science – Arjun College of Technology (2020-07 – 2024-05)