Skip to content
View houcem58's full-sized avatar

Block or report houcem58

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
houcem58/README.md

Houcem Hammami — Technical Manager, AI & Data Engineering

AI & Data systems engineer · Tunisia  |  LinkedIn  ·  Credly  ·  houcem0508@gmail.com


Over seven years I moved from IT Department Head to Enterprise Programme Director to AI Platform Engineering Lead — each step taking on larger teams, more complex systems, and more ambiguous AI problems. I currently lead a 15-person AI & Data Platform team delivering 4 concurrent AI projects, while pursuing applied research at LATICE Laboratory, ENSIT, Université de Tunis on LLM-native ETL architectures. My work spans both sides of the Technical Manager role: the engineering depth to make architecture decisions, and the programme management rigour to deliver on schedule and on budget.


Featured Projects

Real-Time Schema Drift Detection

realtime-streaming-pipeline · Python · Kafka · Docker

Stateful micro-batch drift detector for Apache Kafka streams. Detects 5 drift types (schema rename, type, missingness, value, distribution) validated across 4 public datasets. Includes full evaluation framework with F1/Precision/Recall/FPR reporting, threshold calibration documentation, and two Architecture Decision Records explaining the design choices.

kafka · drift-detection · mlops · streaming · python


Modern Data Platform

modern-data-platform-demo · dbt · Airflow · PostgreSQL · Power BI

4-layer Kimball star schema (staging → dimensions → fact → consumption) with 9 data quality gates (6 source-level + 3 dbt tests), 9-task Airflow DAG with email alerting, Power BI semantic layer with rolling window DAX measures, and 34 unit tests across ingestion and validation modules.

dbt · airflow · postgresql · power-bi · data-engineering


AML Fraud Detection Platform

aml-fraud-detection-platform · GNN · XGBoost · Kafka · FastAPI

Anti-money laundering platform combining a 3-model ensemble — XGBoost tabular scorer, LightGBM behavioural scorer, and HybridGAT graph neural network — with context-aware fusion (ISOLATED vs DENSE transaction network contexts), BCT/ACPR/FinCEN regulatory rule engine, SHAP explainability, and nightly RLHF recalibration with automatic rollback. Full model card and ADR included.

fraud-detection · graph-neural-network · aml · kafka · fintech · regulatory


ConversaETL — Typed Conversational ETL

ConversaETL-Showcase · LLM · Python · Research

Converts natural-language ETL requests into validated, reproducible dataframe transformations via LLM-assisted typed planning + deterministic compiler stack + output contract validation. ConversaBench evaluation: Hybrid Compiler F1=0.964 vs. direct LLM code generation baseline F1=0.308. No LLM-generated code executed at runtime. Research under review at Big Data Mining and Analytics (Tsinghua / IEEE).

llm · etl · typed-compilation · nlp · research · generative-ai


Superstore Analytics — Live Dashboard

superstore-analytics · dbt · DuckDB · Streamlit · Live Demo →

End-to-end serverless analytics pipeline: raw CSV → dbt transformation layer (staging → warehouse → 3 mart models) → DuckDB embedded OLAP → Streamlit interactive dashboard. 8 chart sections, 4 sidebar filters, CI on every push. Zero server setup required.

dbt · duckdb · streamlit · analytics · python


IMDb Sentiment Analysis — Live Demo

imdb-sentiment-analysis · NLP · DistilBERT · HuggingFace · Live Demo →

End-to-end NLP pipeline on 1,273 scraped IMDb reviews: zero-shot labelling, class balancing with contextual augmentation, and 5-model comparison (Logistic Regression → SVM → 1D-CNN → LSTM → DistilBERT fine-tuning). DistilBERT accuracy: 88.6%. Live inference demo via HuggingFace Inference API with VADER fallback.

nlp · distilbert · huggingface · sentiment-analysis · python


Leadership & Programme Delivery

AI Platform Engineering — Management Playbook

ai-platform-engineering · AI Governance · MLOps · Release Management · 2025–Present

Engineering management playbook for a 15-person AI & Data Platform team delivering 4 concurrent AI projects. Covers team structure, AI governance framework with model cards and ethics checklist, MLOps standards (experiment tracking, staged promotion, drift monitoring), blue/green release management at 2 releases/month, and architecture decision records.

ai-governance · mlops · engineering-management · release-management · platform-engineering


IT Department Leadership — Case Study

it-department-leadership · Digital Transformation · Budget Management · Team Leadership · 2017–2020

Management case study for an 8-person IT department: 5 digital transformation initiatives delivered in 3 years, 30% operational cost reduction through infrastructure consolidation and vendor renegotiation, availability raised from 96% to 99.3%, service desk SLA compliance from 72% to 94%. Includes budget breakdown, vendor management framework, and team development approach.

digital-transformation · it-management · budget-management · team-leadership · governance


Agile AI Delivery Playbook

agile-ai-delivery-playbook · TPM · Scrum · Product Owner

Full Scrum delivery playbook from a real programme: AI Intelligence Analytics Platform at Navy HQ (2023–2024). 12-person team, 4 sprints, 30% delivery cycle reduction, 4/4 sprint objectives met. Covers product vision, team/ceremonies, backlog (63 pts across 4 epics), sprint journals with burndowns, velocity analysis, and full release retrospective.

agile · scrum · tpm · product-management · pmp


IT Modernisation Programme — PMO Case Study

it-modernization-program · PMP · PMO · Data Platform

PMO documentation for a real enterprise data modernisation programme (2021–2023): Kafka + Spark streaming ETL, PySpark + dbt + PostgreSQL warehouse, Power BI dashboards. Delivered 2 months ahead of schedule. 99.7% infrastructure availability. Documents cover charter, RACI, WBS, risk register, 3 COPIL steering committee records, and closure report.

pmo · program-management · data-engineering · pmp · governance


Stack

Domain Technologies
Data Engineering dbt · Apache Airflow · Apache Kafka · PostgreSQL · DuckDB · Python
Machine Learning XGBoost · LightGBM · PyTorch Geometric · HuggingFace Transformers · scikit-learn
MLOps SHAP explainability · RLHF · drift detection · model versioning · GitHub Actions
Analytics & BI Streamlit · Power BI · DAX · Kimball star schema · Plotly
Infrastructure Docker · Docker Compose · CI/CD pipelines
Leadership PMP · PSM II · PSPO II · CPMAI · ADRs · Model Cards · Runbooks
Languages Python · SQL (PostgreSQL, dbt-SQL)

Engineering Approach

The technical repos in this portfolio include:

  • ADRs — architecture decisions with rejected alternatives and review triggers
  • Model cards — evaluation metrics, fairness considerations, limitations
  • Runbooks — operational procedures and failure recovery steps
  • Sequence diagrams — system interaction flows in Mermaid
  • Engineering standards — contribution guidelines and quality gates

How I Work as a Manager

working-with-me.md — communication style, decision framework, 1:1 structure, engineering standards, and what I expect from teams.


Open To

Technical Manager / AI Engineering Manager / Data Engineering Manager / Technical Program Manager / AI Platform Lead

Remote · Tunisia · Open to relocation worldwide for senior roles.


Pinned Loading

  1. aml-fraud-detection-platform aml-fraud-detection-platform Public

    AML fraud detection platform: GNN + XGBoost ensemble, Kafka streaming, SHAP explainability, RLHF recalibration. AUC 0.925 on IBM AMLSim. Showcase repo.

    Dockerfile 2

  2. ConversaETL-Showcase ConversaETL-Showcase Public

    Conversational ETL platform: LLM-assisted typed planning + deterministic compiler + output contract validation. Showcase repo. Full implementation private.

    Dockerfile 2

  3. realtime-streaming-pipeline realtime-streaming-pipeline Public

    Real-time schema and statistical drift detection on Apache Kafka streams. 5 drift types. F1=0.955 average across 4 public datasets.

    Python 2

  4. agile-ai-delivery-playbook agile-ai-delivery-playbook Public

    Scrum delivery playbook: AI Intelligence Analytics Platform — TPM + Product Owner, 12-person team, 4 sprints, 30% cycle time reduction. PMP · PSM II · PSPO II.

    1

  5. ai-platform-engineering ai-platform-engineering Public

    Engineering management playbook for a 15-person AI & Data Platform team — governance, MLOps standards, release management, AI ethics framework

    1

  6. it-modernization-program it-modernization-program Public

    PMO case study: Enterprise Data Modernization — ETL platform, Power BI dashboards, data governance. 99.7% availability. PMP-governed delivery.

    1