AI Solutions Engineer | Business-agent deployment, integration, evaluation and observability
Python · SQL · REST APIs · LLM tool-calling · OpenTelemetry · Langfuse · client delivery

I turn operational workflows into testable AI applications. My focus is the full delivery loop: understand the business need, integrate the application with APIs and data, keep sensitive actions behind deterministic controls, evaluate the agent against business-risk scenarios, diagnose failures and iterate.

After fifteen years in client-facing audiovisual delivery, I moved into technical work through applied AI, Python and data projects. I am targeting AI Solutions Engineer, AI Deployment Consultant, Forward Deployed Engineer and Implementation Consultant roles in product teams, consulting firms and IT services companies.

Based in France. Open to remote France/Europe roles and client travel.

Evaluation and model-selection case study for a restaurant workforce-planning assistant.

• 100 business-risk scenarios × 6 configurations = 600 observations.
• Checks across replies, tool calls, APIs and resulting database state.
• Retrieval, mutations, permissions, cross-tenant isolation and prompt injection.
• OpenTelemetry instrumentation exported to self-hosted Langfuse.
• Analysis across quality, write integrity, observed security, p95 latency and cost.
• Failure diagnosis leading to deterministic confirmation and evidence-preservation recommendations—not another prompt-only iteration.

Repository: https://github.com/75016paris/bernardo-agent-evals

Restaurant workforce-planning SaaS demo: web dashboard, API and WhatsApp assistant for schedules, leave and replacement workflows.

The model handles intent, tool selection and language. Deterministic code owns authorization, tenant isolation, confirmations, database writes and staffing constraints.

Demo: https://comptoir.cosmobot.fr
Repository: https://github.com/75016paris/comptoir-ai-restaurant-planning

Internal data tool delivered to a solar company in Bangladesh: multi-site monitoring, photovoltaic forecasting and anomaly detection.

Repository: https://github.com/75016paris/solar-forecasting-anomaly-detection-dashboard

Internal analytics tool delivered to a meal-delivery company in England: conversion, churn, renewals, cohorts and purchase behavior.

Repository: https://github.com/75016paris/subscription-analytics-dashboard

Public repositories use synthetic data to protect client confidentiality.

AI agents     LLM APIs · tool/function calling · prompt engineering · human-in-the-loop
Evaluation    reply/tool/API/database checks · permissions · prompt injection · failure analysis
Observability OpenTelemetry · self-hosted Langfuse · traces · tokens · cost · latency · scores
Integration   REST APIs · JSON · webhooks · auth · multi-tenant boundaries
Python/data   Python · SQL · pandas · NumPy · scikit-learn · Streamlit · Plotly
Delivery      discovery · demos · documentation · troubleshooting · client communication
Ops           Git · GitHub · Linux · SSH · OVH VPS · Caddy · systemd · logs
Web           React · TypeScript · Hono/Bun · Drizzle project experience

I use AI coding tools as an accelerator while keeping ownership of framing, integration, debugging, testing, deployment and documentation.

• AI Solutions Engineer
• AI Deployment Consultant / Technical Consultant
• Forward Deployed Engineer
• Implementation Consultant / Specialist
• Customer Engineer for AI/API products

• Portfolio: https://cosmobot.fr
• Email: info@cosmobot.fr