12 Days to First OilIngénieur Data Management, Réservoir & Production
Ami Ba Mbengue
MSc Data Science & Analytics · Ingénierie Géologique
I am joining PETROSEN E&P in Dakar on 1 September 2026. This is what I need to learn, in what order, and what I will have built before I walk in.
Sangomar · GTAUpstream dataFR / ENApprox. 2h per day
My sprint progress0 / 12 days
Where my data comes from
Everything I will manage starts about two kilometres under the seabed, roughly 100 km off the coast south of Dakar. Oil moves up through subsea wells to a floating production, storage and offloading vessel. Numbers move the same way: pressure, temperature, flow rate, water cut. My role sits at the top of that column, where the numbers land.
Fig. 1 — Subsea wells tied back to a floating production vessel. Every barrel produced generates a row of data somewhere.
Subsurface data
Logs, cores, seismic, formation tops. The rock, described in files.
Production data
Rates, allocation, downtime, well tests. The flow, measured daily.
Standards & governance
OSDU, PPDM, unique well identifiers, data quality, confidentiality.
My toolkit
Python, SQL, Power BI, and the petro-technical software the teams live in.
My 12 day sprint
Twelve stations, one per day, deepest last. I tap a day to mark it complete.
0 m
DAY 01Thu 20 Aug
The company and the asset
PETROSEN Holding vs PETROSEN E&P vs PETROSEN T&S. I need to know exactly which one I am joining and what it owns.
Sangomar: Woodside operates with 82%, PETROSEN holds 18%, first oil June 2024, nameplate 100,000 barrels per day, 23 wells in phase 1.
GTA (gas, operated by bp, on the Mauritania border) and Yakaar-Teranga. I learn the difference between the oil and gas assets in the portfolio.
The DG's stated ambition: PETROSEN wants to become an operator itself, not just a partner. I understand why that raises the stakes on owning our own data.
My deliverableOne page, in my own words, explaining the portfolio to someone who knows nothing about it. If I cannot write it, I do not know it yet.
110 m
DAY 02Fri 21 Aug
Well lifecycle vocabulary, in both languages
The phases: exploration, appraisal, development, production, abandonment. En français: exploration, évaluation, développement, production, abandon.
Well terms: spud, drilling, casing, cementing, completion, perforation, workover, side track, injector vs producer.
Depth references: MD (measured depth), TVD (true vertical depth), datum, deviation survey. I need to know why MD and TVD differ and why confusing them corrupts a database.
Well naming and unique well identifiers. This is the single most common source of chaos in upstream data.
My deliverableA bilingual term sheet I keep open on day one. I start it now and add to it every day.
220 m
DAY 03Sat 22 Aug
Subsurface data types and file formats
Wireline logs: gamma ray, resistivity, density, neutron, sonic, caliper. What each one measures and what a geologist reads from it.
Seismic: 2D vs 3D, pre-stack vs post-stack, SEG-Y files and why they are enormous.
The formats I will handle constantly: LAS, DLIS, SEG-Y, CSV, and vendor exports.
Go deeper — what's actually inside these files
A LAS file is plain text, split into named sections that each start with a tilde. ~V (Version) says which LAS spec applies and whether data wraps to multiple lines. ~W (Well) holds the well name, UWI, field, and the start/stop/step depth of the log. ~C (Curve) lists every mnemonic in the file — GR, RHOB, NPHI, RT — each with its unit and the order it appears in the data block. ~P (Parameter) holds acquisition metadata: tool type, calibration, run number. ~A (ASCII) is the data itself: one row per depth, values separated by spaces, with a sentinel like -999.25 standing in for "no reading."
~C CURVEEvery mnemonic in the file, in column order, with units
~P PARAMETERTool calibration, run number, acquisition metadata
~A DATAThe actual depth-indexed readings, one row per depth
Fig. 2b — Anatomy of a LAS file, top to bottom. This is the header I read line by line on day three.
And here is what one actually looks like — the standard CWLS teaching example, a real (if old) well log from the EDAM field in Saskatchewan, reproduced exactly as written:
~VERSION INFORMATION
VERS. 1.2: CWLS LOG ASCII STANDARD -VERSION 1.2
WRAP. NO: ONE LINE PER DEPTH STEP
~WELL INFORMATION BLOCK#MNEM.UNIT DATA TYPE INFORMATION
#--------- ------------- ------------------------------
STRT.M 1670.000000:
STOP.M 1669.750000:
STEP.M -0.1250:
NULL. -999.2500:
COMP. COMPANY: # ANY OIL COMPANY LTD.
WELL. WELL: ANY ET AL OIL WELL #12
FLD . FIELD: EDAM
LOC . LOCATION: A9-16-49-20W3M
PROV. PROVINCE: SASKATCHEWAN
SRVC. SERVICE COMPANY: ANY LOGGING COMPANY LTD.
DATE. LOG DATE: 25-DEC-1988
UWI . UNIQUE WELL ID: 100091604920W300
~CURVE INFORMATION#MNEM.UNIT API CODE CURVE DESCRIPTION
#--------- ------------- ------------------------------
DEPT.M : 1 DEPTH
DT .US/M : 2 SONIC TRANSIT TIME
RHOB.K/M3 : 3 BULK DENSITY
NPHI.V/V : 4 NEUTRON POROSITY
SFLU.OHMM : 5 RXO RESISTIVITY
SFLA.OHMM : 6 SHALLOW RESISTIVITY
ILM .OHMM : 7 MEDIUM RESISTIVITY
ILD .OHMM : 8 DEEP RESISTIVITY
~PARAMETER INFORMATION#MNEM.UNIT VALUE DESCRIPTION
#--------- ------------- ------------------------------
BHT .DEGC 35.5000: BOTTOM HOLE TEMPERATURE
BS .MM 200.0000: BIT SIZE
FD .K/M3 1000.0000: FLUID DENSITY
MATR. 0.0000: NEUTRON MATRIX(0=LIME,1=SAND,2=DOLO)
MDEN. 2710.0000: LOGGING MATRIX DENSITY
RMF .OHMM 0.2160: MUD FILTRATE RESISTIVITY
DFD .K/M3 1525.0000: DRILL FLUID DENSITY
~Other Note: The logging tools became stuck at 625 meters causing the data
between 625 meters and 615 meters to be invalid.~A DEPTH DT RHOB NPHI SFLU SFLA ILM ILD
1670.000 123.450 2550.000 0.450 123.450 123.450 110.200 105.600
1669.875 123.450 2550.000 0.450 123.450 123.450 110.200 105.600
1669.750 123.450 2550.000 0.450 123.450 123.450 110.200 105.600
Fig. 2c — A real LAS file, unedited. Notice NULL. -999.2500 in the well block — that is the sentinel value used throughout the data section whenever a tool didn't get a reading.
DLIS is the binary cousin: it can hold several runs and tool frames in one file, so it needs a real parser (dlisio), not a text editor. SEG-Y is seismic: a text header, a binary header, then one trace header plus samples for every single trace — which is why a 3D survey file can run into terabytes.
My deliverableI open a LAS file in a text editor, read the header, and understand every line of it.
Fig. 2 — Low gamma ray, high resistivity, high porosity read together mean clean rock holding hydrocarbons. I learn to see this shape at a glance.330 m
DAY 04Sun 23 Aug
Petrophysics refresher
This is my home turf. I speed run it and move on.
Porosity, permeability, water saturation, net to gross, net pay, cut-offs.
Archie's equation and why saturation estimates are always argued about.
How petrophysical interpretation becomes an input to volumes and then to reserves.
Reserves classification: 1P, 2P, 3P, proved, probable, possible. I need to know why reserves numbers are legally sensitive.
My deliverableI can explain, out loud, how a log curve turns into a barrel number on a board slide.
440 m
DAY 05Mon 24 Aug
Reservoir engineering essentials
PVT and fluid properties: bubble point, formation volume factor, GOR, viscosity.
Drive mechanisms and pressure maintenance. Sangomar uses water and gas injection, so I understand why we inject in the first place.
Material balance in concept, recovery factor, and what moves it.
Decline curve analysis: exponential, hyperbolic, harmonic. This is the single most useful chart in my future job.
Go deeper — bubble point and the decline math
Above the bubble point pressure, oil in the reservoir is undersaturated: it is a single liquid phase, and as pressure drops the oil barely changes — GOR stays flat because no gas is coming out of solution yet. The moment reservoir pressure falls below the bubble point, gas starts breaking out inside the rock, free gas appears, and GOR climbs while the oil's formation volume factor (Bo) shrinks. If I ever see a well's GOR suddenly trend upward in the daily numbers, the first question is whether that well's pressure has crossed its bubble point — not whether the meter is broken.
Fig. 3b — GOR against pressure. The kink at Pb is the single most important point on this chart.
Arps' decline equation is q(t) = qi / (1 + b·Di·t)^(1/b). b = 0 gives the exponential line in Fig. 3, b = 1 gives harmonic decline, and anything between is hyperbolic — the b value is really a statement about how much the reservoir's drive mechanism is slowing the decline down.
My deliverableI fit a decline curve in Python on a public production series. Even a rough one counts.
Fig. 3 — Same well, two decline models, very different forecasts. The gap between them is a budget conversation.550 m
DAY 06Tue 25 Aug
Production data and allocation
The daily production report and what each column means. This will land in my inbox every morning.
Allocation: the vessel measures commingled flow, so per-well volumes are back-calculated from well tests. I learn the logic and where it goes wrong.
Water cut, GOR, uptime and downtime coding, deferred production.
Why partners audit these numbers: cost recovery and profit oil under the production sharing contract depend on them.
My deliverableI sketch the allocation chain end to end on one page. Diagram, not prose.
01 · SourceWellhead sensors, subsea meters, well tests
Fig. 4 — The allocation chain. My role touches every box after "measure".660 m
DAY 07Wed 26 Aug
Standards and governance
OSDU: the industry standard subsurface data platform built to break silos. I read the overview docs and understand the idea of a common data model.
PPDM: the Public Petroleum Data Model and the "what is a well" problem. Deceptively deep, genuinely important.
WITSML for drilling, PRODML for production data exchange.
Data quality dimensions: completeness, accuracy, consistency, timeliness, uniqueness. Master data vs transactional data. Metadata and lineage.
Go deeper — what each standard actually governs
WITSML is real-time: an XML feed that streams drilling parameters — depth, mud weight, ROP — from the rig floor to the office while the well is still being drilled. PRODML is its production-side sibling: it defines how production volumes, well tests, and allocation data move between systems once the well is on stream. PPDM sits underneath both — it is the reference data model that defines, precisely, what a "well" is, so that a well in the drilling system, the production system, and the reserves system all point to the same record instead of three slightly different ones. OSDU is the newest layer: a cloud data platform that tries to hold subsurface, well, and seismic data under one common schema so any application — Petrel, a Python notebook, a partner's tool — can read it the same way.
OSDUCloud platform, common schema, sits on top so any application can read the same data
WITSMLReal-time drilling data, rig floor to office, while the well is being drilled
PRODMLProduction data exchange — volumes, well tests, allocation — once the well is on stream
PPDMThe reference model underneath all of it: what a "well" is, canonically
Fig. 5b — Four standards, four different jobs. PPDM is the foundation the other three quietly depend on.
My deliverableI can say what OSDU is in two sentences without notes. This alone will make me sound serious on day one.
ConsumptionDashboards, reports, reservoir and production models
GovernanceStandards, naming rules, ownership, quality checks, access control
StorageCommon data model, master well register, document store
IngestionLAS, DLIS, SEG-Y, daily reports, vendor and partner deliveries
Fig. 5 — Read it bottom up. Most upstream data problems are created at ingestion and paid for at consumption.770 m
DAY 08Thu 27 Aug
The tool stack
I will not have licences yet. I learn the names, the purpose and the vocabulary so nothing in a meeting is a black box.
PetrelSubsurface modelling and interpretation, the centre of gravity for geoscience teams
TechlogPetrophysics and log analysis
OFM / Energy ComponentsProduction data management and surveillance
Eclipse / tNavigatorReservoir simulation, where the forecasts come from
Pythonpandas, lasio, welly, matplotlib. My actual daily weapon
SQL + Power BIQuery the store, then make people believe the answer
My deliverableI install Python with lasio and pandas tonight, so day 9 starts with code and not setup.
880 m
DAY 09Fri 28 Aug
Hands on, part one: well data
I use the Volve field dataset, released publicly by Equinor, or another open well data source. Then:
Load a LAS file with lasio and inspect the curve mnemonics and units.
Run quality checks: null values, out of range readings, mismatched units, duplicate depths.
Plot a three track log display.
Build a clean well header table: well name, unique identifier, coordinates, total depth, spud date.
My deliverableA notebook that turns a raw LAS file into a clean, documented table. This is literally the job.
990 m
DAY 10Sat 29 Aug
Hands on, part two: production data
Take a public production history and build monthly rates per well.
Calculate water cut and GOR over time, flag anomalies and gaps.
Fit a decline curve and produce a simple forecast.
Wrap it in a one page dashboard. Plain and legible, not decorated.
My deliverableA small portfolio piece I can show my supervisor in week one. Two notebooks beat any certificate.
1100 m
DAY 11Sun 30 Aug
The rules around the data
Production sharing contracts in outline: cost oil, profit oil, state share. Why data accuracy is money.
Confidentiality. Subsurface and production data are commercially sensitive and often state sensitive. I assume nothing leaves my laptop, ever, including into any external tool.
Reporting obligations: the ministry, and ITIE Senegal for transparency reporting.
Local content policy and what PNICL is trying to achieve. I am part of a deliberate national skills strategy, so I know the argument.
HSE basics and offshore induction vocabulary, in case I am ever offered a site visit.
Go deeper — where the barrels actually go
Under a production sharing contract, gross production is not split evenly from day one. The contractor first recovers its exploration, development, and operating costs out of a capped share of production called cost oil — the cap is set in the contract, often somewhere around 40–60% of gross, and varies by field. Whatever is left after cost recovery is profit oil, and that is what gets divided between the contractor and the state, often on a sliding scale tied to production rate or to an R-factor — the cumulative ratio of revenue to cost — so the state's share tends to rise as the field becomes more profitable. PETROSEN sits on both sides of that split: as an 18% equity partner in Sangomar it earns a working interest share directly, and separately, as the state's data steward, the numbers I help produce are what the ministry and JV partners audit to check the cost oil and profit oil split is being calculated correctly.
Gross production, 100%
Cost oil — cost recoveryProfit oil
Profit oil split
Contractor shareState share (incl. PETROSEN)
Fig. 6 — Illustrative split only, actual percentages are contract-specific and confidential. The mechanism, not the numbers, is what matters here.
An allocation error is not just a data problem here — it can move real money between the contractor and the state.
My deliverableMy personal rule for what I will never paste into an external tool. I decide it now, not under deadline pressure.
1210 m
DAY 12Mon 31 Aug
Land the first week
Print my bilingual term sheet. It stays in my bag.
Prepare a two minute answer to "parlez-nous de votre parcours" that connects geology to data science on purpose, not by accident.
Write my questions for my supervisor (list below).
Set a personal goal for week four: one thing I will have improved, however small. A cleaner well register is a legitimate answer.
My deliverableSleep. Arrive early. Say little, take notes, ask precise questions.
My unfair advantage
Most people in an upstream data role arrive from one side only. Geoscientists who cannot code, or data people who do not know what a formation top is and quietly average across two different reservoirs.
I have geological engineering and a data science masters. I can read the rock and write the pipeline. I do not present myself as an intern learning oil and gas. I present myself as someone who already speaks both languages and is learning where they meet in this specific company.
Glossary, FR / EN
Term
What it means
Français
Upstream
Exploration and production, before transport and refining
Amont
Well log
Continuous measurement of rock properties against depth
Diagraphie
Core
Physical rock sample cut from the well
Carotte
Formation top
Depth where a geological unit begins
Toit de formation
Net pay
Thickness of rock that will actually produce
Épaisseur utile
Porosity
Void space fraction in the rock
Porosité
Permeability
Ability of the rock to let fluid flow
Perméabilité
Water cut
Share of water in total liquid produced
Taux d'eau
GOR
Gas to oil ratio
Rapport gaz-huile (RGH)
Allocation
Splitting measured total flow back to individual wells
Allocation de production
Well test
Measuring one well's rate in isolation
Test de puits
Workover
Intervention on an existing well
Reconditionnement
Injector
Well used to inject water or gas, not to produce
Puits injecteur
Recovery factor
Share of oil in place that can actually be produced