Expertise

Ground-up, end to end.

I didn't arrive at AI platforms from a framework tutorial — I built up to them: logs, ETL, data lakes and BI first, then model serving, retrieval, evaluation and agents on top. That arc is why the platform holds together.

The arc

2011 – 2017

Analytics & BI — RR Donnelley, Chennai

Market intelligence, forecasting and statistical analysis for international clients; leading analyst teams and learning what decision-makers actually need from data.

2017 – 2018

MSc Big Data Analytics — IÉSEG, France

The pivot to Europe and to engineering-grade data work.

2019 – 2023

Data engineering foundations — international organisation, Lyon

ETL and log pipelines (NiFi, MiNiFi, Talend) feeding HDFS/Hive data lakes and search analytics (Elasticsearch, OpenSearch, Solr, Kibana); dashboards and KPIs on top.

2023

First LLM chatbot — on CPU

Pioneered the organisation's first internal chatbot, benchmarking open-source LLMs and defining the GPU/VRAM requirements that justified real AI infrastructure.

2024 – present

AI platform ownership

Enterprise AI cluster across multi-server multi-location GPU infrastructure: vLLM serving behind a LiteLLM proxy, with Graph RAG, specialised OCR/translation/transcription/vision services, MCP-based agentic workflows, automated fine-tuning cycles — with MLflow evaluation and Prometheus/Grafana monitoring throughout.


The map

Drag, zoom, and hover to trace how the pieces connect. The gold dashed path is the career evolution above.


The stack

AI / LLM engineering

vLLMLiteLLM LlamaIndexRAGFlow BGE-M3KG-RAG MCP / agentic AILoRA / PEFT fine-tuning prompt engineering

MLOps & evaluation

MLflowPrometheus Grafanamodel benchmarking evaluation pipelinesSOPs & runbooks

Backend & orchestration

PythonFastAPI Apache NiFiAirflow n8nTalend

Data & infrastructure

CUDA / NVIDIAJupyterHub PostgreSQLHDFS / Hive Elasticsearch / OpenSearchQlikSense