[01] IPSOS France · 2022–present
Autonomous Analytical Reporting Agents
@ipsos_france
// problem
Manual extraction and synthesis of data from hundreds of PowerPoint presentations weekly was a critical bottleneck for enterprise research workflows.
// solution
Designed multi-agent orchestration systems using LangChain and LlamaIndex to automate end-to-end analytical pipelines from unstructured inputs to publishable outputs.
→ Full pipeline automation; hours of manual work eliminated.
LangChain
LangGraph
LlamaIndex
Azure OpenAI
Docker
[02] IPSOS France · 2022–present
RAG Pipeline & Vector Store Optimisation
@ipsos_france
// problem
LLM retrieval over large, heterogeneous enterprise corpora (BigQuery + unstructured docs) was slow and imprecise, degrading generation quality.
// solution
Architected RAG systems with Qdrant and FAISS vector stores; optimised chunking, re-ranking, and embedding strategies. Implemented LLMOps monitoring with LangSmith and MLflow.
→ +40% text-processing efficiency improvement.
RAG
Qdrant
FAISS
LangSmith
MLflow
[03] IPSOS France · 2022–present
Text-to-SQL Agent for Natural Language DB Queries
@ipsos_france
// problem
Non-technical stakeholders at a global research firm were entirely dependent on data teams for any ad-hoc database query — creating bottlenecks across business units.
// solution
Built a conversational Text-to-SQL agent translating natural language into optimised BigQuery SQL, integrated into automated reporting flows and self-service dashboards.
→ Self-service data access for non-technical users.
LangChain
BigQuery
GPT-4
Vertex AI
GCP
[04] IPSOS France · 2022–present
Multimodal Brand Visibility & Sentiment Analytics
@ipsos_france
// problem
Media intelligence clients needed scalable brand tracking across thousands of images, videos, and documents — impossible at speed with manual review processes.
// solution
Deployed Vision-Language pipelines (GPT-4V) for automated brand detection, OCR, and multimodal sentiment analysis over large media asset libraries on GCP Vertex AI.
→ Enterprise-scale automated brand analytics.
GPT-4V
YOLO
OCR
Vertex AI
BigQuery
[05] Henceforth · Mar–Sep 2021
Edge AI Optimisation & Generative Pipelines
@henceforth_rabat
// problem
Generative models (Text-to-Image, ASR) needed to run reliably on resource-constrained hardware while maintaining production-quality outputs.
// solution
Implemented Hugging Face diffusion pipelines for image generation and a custom deep-learning ASR system with DSP. Applied INT8 quantisation, ONNX export, and TensorRT optimisation.
→ −30% inference latency for edge deployments.
Hugging Face
TensorRT
ONNX
INT8
DSP
[06] IPSOS France · 2022–present
Enterprise GenAI Platform Architecture
@ipsos_france
// problem
POC-stage AI models needed to be industrialised into secure, compliant, monitored enterprise platforms — bridging data science, ML engineering, and business stakeholders.
// solution
Designed cloud-native AI platforms on Azure AI Foundry and GCP Vertex AI with LLMOps pipelines (MLflow, LangSmith), RBAC/IAM security, GDPR compliance, and CI/CD for AI services.
→ Production-grade, GDPR-compliant AI platform at enterprise scale.
Azure AI Foundry
LangGraph
AutoGen
MLflow
Kubernetes
