Currently · CEPS, Paris

Alex Komla LabouData & AI Engineer.

I connect data, models and business needs to build AI systems that can be checked.

Data engineering, machine learning and AI agents — statistics as a foundation, evaluation as a requirement.

See my work at CEPSExplore projects
AL
Alex Komla Labou
Data · ML · Agentic AI
9
Education
ENSAI Rennes · ENSAE Dakar
Focus
Data · ML · Agentic AI
Languages
French · English
PythonSQLDuckDBBigQuerydbtFastAPINext.jsTypeScriptPyTorchLightGBMLangGraphHugging FaceFAISSGCPCloud RunTerraformDockerGitLab CI
01 — Professional work · June 2026 — present

A complete platform. Every figure traceable.

At CEPS, I am the sole designer and developer of a data platform and its business application, working with domain teams and the ministerial IT department (DNUM).

CEPS Explorer · amounts, dates and sensitive information masked
CEPS Explorer: medicine economic data dashboard and source tracking. Amounts and dates masked.

CEPS Data Platform& CEPS Explorer

PythonDuckDBFastAPINext.jsTypeScriptSQLGitLab CI

Medicine-price negotiations require sales, reimbursement, hospital, rebate and regulatory data. The starting point: scattered files and competing identifiers.

I built a pipeline that validates these sources, preserves their history and brings analysis into one application.

integrated sources
database rows · 6.4 GB
business & quality SQL views
automated tests

Figures as of 30 September 2026

An application for business teams

Product search, sales and reimbursement by channel, price/volume/mix growth decomposition, HAS assessments and exceptional-access pathways, CSV/Excel exports and one-page product PDFs.

End-to-end traceability

442 source files tracked by SHA-256, 675 logged ingestion runs and 3,922 recorded data-quality issues. Preserved sources, transactional imports and human-approved corrections.

  1. 01SourcesExcel · CSV · PDF
  2. 02IngestionPython
  3. 03WarehouseDuckDB · 65 tables
  4. 04APIFastAPI · read-only
  5. 05ApplicationNext.js · Explorer
Case study
02 — Projects, internships & research

From prediction to decisions.

End-to-end projects, where models meet reliability, deployment and usability constraints.

01 / DATA → DECISIONGCP validation

RetailOps AI

A demand forecasting platform that turns M5 data into inventory scenarios, explanations and recommendations for human decisions.

BigQuery · dbtLightGBM · SHAPLangGraph · GeminiCloud Run
0.771Mean bottom-level RMSSE · 28-day validation
02 / UN WFP · BIOMETRICSWorld Food ProgrammeInternship · UN World Food Programme

WFP Biometric Deduplication

An in-house fingerprint deduplication prototype so that every humanitarian aid beneficiary is registered only once: deep-learning minutiae extraction (MinutiaeNet) and Bozorth3-inspired matching, ported to Python.

PythonMinutiaeNet · CNNBozorth3 · graph matchingSOCOFing
4.34%EER in 1:1 verification · AUC 0.981 · Rank-1 94.5%
03 / EVIDENCE-GROUNDED AIPrivate beta

JobCopilot

An application assistant that links every claim to verified profile evidence, then prepares a reviewable, human-supervised application pack.

LangGraphFAISS · PydanticStreamlitGmail · Calendar
121/121supported claims · frozen 50-case benchmark
An application workflow
  1. 1Human-verified profile
  2. 2Relevant evidence retrieval
  3. 3Matching & evidence ledger
  4. 4Draft reviewed before action

Facts come from the profile. The model selects the evidence.

04 / RELATION INTELLIGENCEENSAIV1 model evaluated

RelationScope NLP

Identify relations between entities, calibrate confidence and abstain on unknown relations. Extended from ENSAI work with Maxime Ferret.

ModernBERTPyTorch · FastAPINext.jsGCP · Terraform
89.16%Macro-F1 · known-relation test in the 56-known open-set protocol
Illustrative example · relation extraction

Marie CurieHEAD discovered poloniumTAIL.

open-setKNOWNUNKNOWN → abstain

A calibrated decision, with the option to abstain.

05 / AGENT EVALUATIONResearch protocol

EviSuff Finance

Does an agent change its answer when necessary evidence disappears? A counterfactual protocol for evidence-sufficiency evaluation in long financial documents.

Counterfactual evaluationAbstention · calibrationPython · CI
4paired conditions: full, gold-only, necessary removal, irrelevant removal
One question, four contexts
FULL

All evidence

GOLD ONLY

Sufficient evidence only

NECESSARY REMOVAL

Critical evidence missing

IRRELEVANT REMOVAL

A distractor removed

Protocol schematic · no empirical model results claimed.

06 / CLOUD-TO-EDGE · RLENSAIIRISAFinal-year project · ENSAI × IRISA

Smart Bidding

Where should a serverless function run across cloud, fog and edge? A reverse Vickrey auction (GIRAFF), greedy heuristics and a Dueling Double DQN agent compared in the FaaS-Sim simulator. Team project with Richard Gozan and Lynn Siewe.

Python · FaaS-SimReverse Vickrey auctionDueling Double DQNMDP · action masking
100%SLAs accepted by the DQN agent, vs 92.8% for GIRAFF · 27-node topology
Also
IMDB Data Platform

TSV → Parquet → GCS → BigQuery ingestion, chunked processing and a reproducible bronze layer.

View source
03 — Experience

Strong foundations. Demanding contexts.

From medicine data to fingerprints: working with business context, real constraints and explicit evaluation criteria.

  1. June 2026 — present
    Paris, France
    CEPS

    Data Science & Data Engineering · Internship

    CEPS · Economic Committee for Health Products (France)

    Sole developer of the CEPS Data Platform and CEPS Explorer: ingestion from 15 sources, history, quality controls, economic analysis, a FastAPI service and Next.js interface. Internal deployment planning with DNUM.

    Python · DuckDB · FastAPI · Next.js · SQL · GitLab CI
    Case study
  2. July — Dec. 2025
    Rome, Italy
    World Food Programme

    Data Scientist · Internship

    World Food Programme · United Nations

    Fingerprint duplicate-detection prototype for beneficiary identification. A hybrid pipeline combining MinutiaeNet minutiae extraction with Bozorth3-inspired matching, including threshold and robustness analysis.

    SOCOFing · EER 4.34% · AUC 0.981 · Rank-1 94.5%
    Case study
  3. Apr. — June 2024
    Dakar, Senegal
    CES

    Research Assistant

    Cabinet des Experts Statisticiens

    Research methodology support, data cleaning and analysis, result interpretation and statistical report structure.

  4. Sept. — Oct. 2022
    Dakar, Senegal
    ANSD

    Data Analyst · Internship

    ANSD · Senegal National Agency of Statistics and Demography

    Exploratory analysis and survey-data quality checks. Studied sociodemographic migration determinants using logistic regression.

Education
ENSAI

ENSAI Rennes

2024 — 2026

Data Science & Data Engineering

Machine learning, deep learning, NLP, cloud and data engineering.

ENSAE Pierre Ndiaye, Dakar

ENSAE Dakar

2020 — 2024

Engineering degree in Statistics & Economics

Statistics, econometrics, inference and survey analysis.

04 — How I work

Trust is engineered.

01 / DATA

A foundation you can check.

Structure sources, define contracts and trace transformations before asking a model to produce an answer.

02 / EVALUATION

A result with its context.

Compare with baselines, separate calibration from testing and document errors. A metric is meaningful only with its protocol.

03 / PRODUCT

A supervised decision.

Give users evidence, limitations and understandable tools. Keep human review where the decision requires it.

Toolkit

AI / ML / NLP

PythonPyTorchTensorFlowscikit-learnHugging FaceLightGBMLangGraphFAISSPydantic

Data / Cloud / Delivery

SQLDuckDBBigQuerydbtFastAPINext.jsDockerGCPAWSTerraformGitHub ActionsR Shiny
05 — Contact

Let’s build what comes next.

Applied AI, machine learning, agentic systems and evaluation-oriented research. Get in touch directly.

alexlabou03@gmail.com