Oli Niloy

AI Engineer who ships production LLM systems end to end — multi-agent architectures, RAG pipelines, real-time voice AI, and fine-tuned open-source models. Delivered platforms serving healthcare, financial decision systems, and construction clients, including a 5,050-persona conversational system and a sub-500ms real-time meeting co-pilot. Strongest in Python, FastAPI, vLLM, and the practical engineering that gets a model from notebook to a low-latency service people depend on.

AI/ML Engineer (promoted from Trainee, Jul 2025) at Join Venture AI (2025-05 – Present)

Designed and deployed production AI systems for international healthcare, finance, enterprise automation, and construction intelligence clients.

• Architected RAG pipelines and vector search across FAISS, Pinecone, and Qdrant for retrieval over unstructured domain corpora.
• Fine-tuned open-source LLMs (Qwen family) with LoRA and QLoRA on custom datasets, cutting one training cycle from ∼73 hours to ∼2 hours.
• Built multimodal pipelines combining custom OCR, computer vision (Google Vision API), and LLM reasoning.
• Built real-time systems using streaming speech recognition, WebSockets, multi-agent orchestration, and low-latency LLM inference.
• Shipped scalable inference APIs on FastAPI + vLLM, running GPU-heavy workloads with health checks and zero-downtime reloads.

B.Sc. in Computer Science and Engineering – East West University (2025)

Higher Secondary Certificate (HSC) in Science – Dhaka Imperial College (2019)

Secondary School Certificate (SSC) in Science – Rajdhani Ideal School (2017)