Applied AI Engineer · NLP · Multilingual AI

I turn real problems into reliable AI products.

LLM applications, intelligent automation, multilingual NLP, structured data workflows and production-minded AI systems.

Computer Engineering NLP / Siri experience French · English · Hebrew Python · APIs · SQL

Selected systems

Not demo chatbots. Each project is built around a real workflow, explicit constraints, measurable quality and clear failure handling.

01 · Flagship project · In progress

Hebrew WhatsApp Accident Intake Assistant

Automates first-line intake after a car accident while keeping the workflow controlled: required questions, validated fields, approved answers, structured storage and automatic human escalation when the system is uncertain.

PythonFastAPIPostgreSQLWebhooksStructured outputHuman-in-the-loopDocker-ready

What this project proves

API/webhook integration and conversation state
LLM extraction into a strict schema instead of free-form generation
Validation, fallbacks, retries and safe handoff
Portable architecture designed to move to Google Cloud without rewriting the product

02 · Multilingual AI

Multilingual RAG Assistant

A grounded assistant for French, English and Hebrew documents, designed around retrieval quality, citations and reproducible evaluation.

RAGEmbeddingsVector DBCitationsEvals

03 · AI quality

LLM Evaluation Lab

A small evaluation harness for testing extraction quality, regressions, latency and cost — because production AI needs evidence, not “looks good to me”.

Golden setsRegression testingLatencyCost

04 · Creative AI

Creative AI Studio

Prompt-driven experiments across music and video, documented as an engineering workflow: constraints, iterations, model behavior, failures, revisions and final output.

Prompt engineeringMultimodal AIIterationCreative toolingEvaluation

Why it belongs in an engineering portfolio

Shows deliberate prompt iteration rather than one-shot generation
Documents failure modes and model limitations
Demonstrates product taste, experimentation and multimodal thinking

What I can prove

The portfolio is organized around evidence: systems, architecture, tests, decisions and outcomes.

LLM application engineering

Structured outputs, prompt design, constrained workflows and safe fallbacks.

Backend + data

FastAPI, REST/webhooks, PostgreSQL schemas, validation and stateful workflows.

Evals + reliability

Golden sets, regression checks, idempotence, retries, error handling and handoff.

Product thinking

From business requirement to scoped V1, measurable quality and production path.

How I build

Clear scope first, then engineering, evaluation and deployment — with the human workflow always visible.

01 · Frame

Define the job

Clarify the user, workflow, data, constraints and failure cases before adding AI.

02 · Build

Make it work

APIs, schemas, state, storage and a controlled end-to-end path.

03 · Evaluate

Measure quality

Test sets, accuracy, latency, cost, edge cases and regression behavior.

04 · Ship

Design for production

Logging, fallbacks, privacy, deployment, observability and human override.

Background

NLP / Product AI

Apple — Siri

Experience working on language technology and NLP in a production product environment.

Education

Computer Engineering — Technion

Technical foundation across software, systems and engineering problem solving.

Current focus

Applied AI systems

Building portfolio projects around LLM applications, multilingual NLP, RAG, evals, automation and production-oriented backend systems.

Let’s build useful AI.

I’m interested in Applied AI, LLM engineering, multilingual NLP and AI product roles where strong product thinking matters as much as model capability.

joyce.bessis@gmail.com