Skip to content

Hello — I'm Abdessalam Bahafid.

Lead Full Stack Data Engineer & AI Architect

I build voice AI & data platforms that hold up at national scale.

I engineer cloud-native AI platforms, distributed real-time data pipelines, and production LLM & Voice systems across AWS and Azure — taking mission-critical projects from architecture to national-scale deployment.

Explore my work Let's talk

Voice pipeline playground

Place a call, then flip the optimization on and call again.

  1. Listen
    VAD endpointing
    1.9s
  2. STT
    streaming
    1.3s
  3. LLM + RAG
    orchestration
    2.6s
  4. TTS
    neural voice
    1.6s

Caller…

Agent

0.0s / 7.4s end-to-end

Totals from a production IVR · Genesys CRM (7.4s → 6.1s). Stage split and dialogue are illustrative.

About

Production-first data & AI engineering.

For 4+ years I've led platforms from the first architecture sketch to production: an enterprise multi-tenant conversational AI SaaS (HolysticX) and the official AI voice assistant of the Africa Cup of Nations, CAN Morocco 2025.

In parallel, I'm a Ph.D. candidate researching Large Language Models for Moroccan Darija, and an Adjunct Professor of Big Data Engineering and Data Warehousing at INSEA.

Abdessalam Bahafid — illustrated portrait
Now
Lead Full Stack Data & AI Engineer at ToumAI
Based in Rabat, Morocco
--:--:--
abdessalam@pf:~

❯ whoami

Abdessalam Bahafid

Lead Full Stack Data Engineer & AI Architect

❯ cat focus.txt

Voice AI

Real-time STT → LLM → TTS pipelines in Darija, French, English and Swahili.

Data pipelines

Event-driven and scheduled ETL/ELT — from on-prem PII stages to cloud warehouses.

Cloud & security

Terraform-managed AWS and Azure under least-privilege IAM and zero-trust networking.

Research & teaching

Ph.D. on LLMs for Moroccan Darija; adjunct professor of Big Data at INSEA.

❯ uptime

4+ years shipping production AI & data platforms

years shipping production AI & data platforms
4+
languages in real-time Voice AI — Darija, French, English, Swahili
4
F1-score on production churn prediction
95%
engineering students mentored at INSEA
100+

Selected work

From architecture sketch to production.

Seven systems, one line each. Open any card for the full case study.

  • Voice AI

    HolysticX

    Multi-tenant SaaS to build, deploy and monitor multilingual voice and chat agents.

    AWS EKSECSLambdaRDS

  • Voice AI

    CAN Voice

    The official multilingual voice assistant of the 2025 Africa Cup of Nations.

    FastAPILangChainPostgreSQLRedis

  • Voice AI

    IVR Churn Platform

    Production IVR on Genesys CRM: churn prediction and faster voice responses.

    Azure FunctionsAzure OpenAIFastAPILangChain

  • Data Engineering

    Hybrid Banking PII Pipeline

    On-prem PII de-identification feeding a cloud ETL engine for banking KPIs.

    On-Prem / Cloud Hybrid ETLPostgreSQLData Warehousing

  • NLP

    ESG Sentiment Engine

    Transformer fine-tuning for French and Wolof ESG sentiment, scheduled on Airflow.

    PyTorchHugging Face TransformersApache Airflow

  • Cloud & Security

    Multi-Cloud Platform & Zero-Trust

    Terraform-automated AWS and Azure environments under least-privilege, zero-trust IAM.

    TerraformGitLab CI/CDAWS IAMCognito

  • Data Engineering

    Parliamentary Data Warehouse

    ETL and tuned PostgreSQL schemas for national parliamentary activity data.

    PostgreSQLETLData Warehousing

  • The full picture

    Dates, scope and every system in one place.

Career

Building in industry, teaching in academia.

  1. Jan 2023 — Present

    ToumAI · Rabat · Hybrid

    Lead Full Stack Data & AI Engineer

    • Architect and lead HolysticX, ToumAI's enterprise conversational AI & voice analytics SaaS.
    • Architected and deployed CAN Voice, the official AI voice assistant of CAN Morocco 2025.
    • Own multi-cloud infrastructure, CI/CD, and zero-trust platform security across AWS and Azure.
    • Deliver NLP and data-engineering systems for telecom, banking, ESG, and public-sector clients.

  2. Feb 2024 — Present

    INSEA · Rabat

    Adjunct Professor · Data Engineering

    • Teach graduate engineering students the Big Data and data-warehousing stack I use in production — see Teaching.

Research

Teaching machines to speak Darija.

Ph.D. · INSEA · 2023 —

Development and Optimization of General Large Language Models for Moroccan Darija.

LLM Pre-trainingFine-TuningLow-Resource NLPTokenization
Publication · Springer2025

Machine Learning and NLP-Based Approach for Constraint Acquisition Problems

A. Bahafid, Z. Erraji, N. Abdou Mohamed, A. Allak, K. Gaanoun, I. Benelallam

Systems Modelling and Simulation · Springer Nature Singapore · pp. 146–158

Abstract

Integrates Natural Language Processing with constraint acquisition through a conversational agent, so novice users can describe NP-complete problems — planning, scheduling, resource optimization — in natural language. The system interprets the description, formulates it as a constraint network, solves it, and presents alternate solutions when available.

doi:10.1007/978-981-96-4613-5_11

Teaching

Teaching the stack I ship.

Adjunct Professor · Data Engineering at INSEA, Feb 2024 — Present.

INSEA — National Institute of Statistics and Applied Economics

100+

graduate engineering students coached

  • Big Data ecosystem

    Hadoop / Cloudera — HDFS, MapReduce, Hive, Pig

  • Data warehousing

    Automated production ETL/ELT with Airflow, dbt and PostgreSQL

  • Business analytics

    Dashboards and exploration with Apache Superset

  • Knowledge discovery

    Advanced laboratory courses and practical workshops

The lab pipeline students build

  1. HDFS
  2. Hive
  3. Airflow
  4. dbt
  5. PostgreSQL
  6. Superset

Toolbox

The stack behind the systems.

Filter the stack by domain.

Showing 55 skills

  • PyTorch
  • Hugging Face
  • LangChain
  • LangGraph
  • LlamaIndex
  • vLLM
  • Unsloth
  • Ollama
  • Apache Spark
  • Apache Airflow
  • Kafka
  • dbt
  • Databricks
  • Hadoop
  • Hive
  • AWS
  • Azure
  • GCP
  • Docker
  • Kubernetes
  • Terraform
  • GitLab CI/CD
  • GitHub Actions
  • NetBird
  • FastAPI
  • Celery
  • PostgreSQL
  • Redis
  • MongoDB
  • Qdrant
  • ChromaDB
  • Weaviate
  • PL/SQL
  • Python
  • Scala
  • SQL
  • TypeScript
  • Bash
  • Java
  • C/C++
  • R
  • PHP
  • Rust
PracticesRAG / RAGASPrompt EngineeringModel EvaluationFine-TuningETL / ELTData WarehousingREST APIsMicroservicesEvent-DrivenDistributed SystemsHigh AvailabilityMulti-Tenant SaaS

Education & languages

Foundations.

  1. 2023 — Present

    Ph.D. Candidate · Computer Science / AI

    INSEA — National Institute of Statistics and Applied Economics

    LLMs for Moroccan Darija: pre-training, fine-tuning, low-resource NLP, tokenization.

  2. 2020 — 2022

    M.Sc. · Information Systems & Intelligent Systems

    INSEA — National Institute of Statistics and Applied Economics

    Machine Learning, NLP, Big Data. Thesis on ML/NLP conversational approaches for automated Constraint Acquisition.

  3. 2017 — 2020

    B.Sc. · Software Engineering

    Ibn Zohr University, Faculty of Sciences — Agadir

    Algorithms & data structures, object-oriented design, relational databases, distributed applications.

Spoken languages

  • AmazighNative
  • ArabicNative / Bilingual
  • EnglishFull Professional
  • FrenchFull Professional

Contact

Let's build something that ships.

Open to collaborations in data engineering, voice AI and LLM systems.

Send me a message

© 2026 Abdessalam BahafidSelf-hosted on Oracle Cloud · Caddy · Cloudflare Tunnel

Quick human check

My email and CV are kept away from scrapers. Confirm you're human and they're yours.