An estimate without an uncertainty interval is a promise nobody knows how to keep. This subject combines machine learning and operations research to produce honest project plans — and revise them when reality drifts.
Context
At B2H, splitting a client project into tasks, estimating durations and assigning the right people is still done by hand, and approximately. Delays are discovered late and replanning is expensive. The goal is a tool that proposes a realistic plan, quantifies its risk and updates it as the project moves forward.
What you will do
- Forecast task durations with uncertainty intervals (quantile regression, conformal prediction)
- Assign tasks through constraint optimisation (CP-SAT, OR-Tools): skills, availability, priorities, deadlines
- Simulate delay risk with Monte-Carlo methods
- Implement incremental replanning that adjusts the existing plan without overturning it
- Automatically break a requirements document into estimated tasks using an LLM
- Surface the result in a kanban and sprint view
Technical challenges
- Cleanly coupling operations research and machine learning: the model's uncertainty must feed the optimiser
- Keeping solve times compatible with interactive use
- Producing calibrated intervals from a limited project history
- Making the plan understandable and editable by a project manager
Deliverables
- A working prototype integrated into the B2H platform
- A forecasting module with calibrated intervals, and its evaluation
- An assignment and replanning module
- Technical documentation and a demo
- The final-year report
Success criteria
- Duration forecast error measured, with empirical interval coverage matching the target level
- Generated plan quality compared with a manual plan on real cases
- Solve time measured as a function of project size
- Replanning that avoids unnecessary reassignments
Tech stack
- OR-Tools CP-SAT
- Conformal prediction, quantile regression
- LLMs
- Python, PostgreSQL
- Kanban, sprints
Who we are looking for
- A final-year engineering or master's student in computer science or data
- An appetite for optimisation and mathematical modelling
- Solid Python
- Product sense: a useful plan is one people understand
- Fluent French and technical English
Practical details
- Duration: 4 to 6 months (February – June 2027, indicative)
- Supervised by the B2H engineering team, with a weekly check-in
- Integrated into the internal B2H platform (single identity, event bus, shared data)
- How to apply: CV + cover letter, at most 2 subjects per candidate
- Deadline: 15/12/2026