À propos de ce poste Data Scientist III (Team- Valmo) chez Meesho
About us
Meesho - Every story here begins with a spark of ambition and a touch of entrepreneurial magic. We're not just a platform — we're your partner in transforming dreams into reality.
Our Mission
Democratising internet commerce for everyone - Meesho started with a single idea in mind: to be an e-commerce destination for the next billion Indian consumers and enable small businesses to succeed online. We provide sellers with a range of industry-first benefits such as zero commission and affordable shipping costs, growing their business by tapping the company's massive customer base, state-of-the-art tech infrastructure, and pan-India logistics at cost through third-party logistics providers and our in-house logistics arm Valmo, in an 'Everyday Low Cost' channel for sellers.
Affordable, relatable merchandise mirroring local markets has helped us make inroads with first-time internet users in the country. We cater to an underserved and unique customer base and cover every serviceable pincode in the country. Our unique business model and continuous innovation have enabled us to become the first Indian horizontal e-commerce company.
Know more about Meesho here: https://www.meesho.io/
About Valmo and the Data Science Team
Valmo is Meesho's own logistics arm, built to give sellers and customers cutting-edge, low-cost logistics and supply chain support. Supply Chain Network Optimization sits at the heart of the Valmo Data Science team: we design and plan the network that moves Valmo's shipments every day, from first-mile pickup through sort centres and line haul to last-mile delivery. We use OR, MILP, stochastic optimisation, simulation, ML and DL across network design, first/middle/last-mile optimisation, capacity planning and delivery promise, and we build for the long term, not just today's problems. This role is for someone ready to execute at speed, with a growth mindset, to build and scale these models.
About the Role
As a Data Scientist 3(Operations Research), you will own a workstream within Valmo network design — not just a model. You will be accountable for the North Star Metric of that workstream (cost per shipment, end-to-end shipment speed, network capacity utilisation, on-time delivery, delivery promise), set its OR/ML roadmap, and mentor DS1s and DS2s in delivering it. You will take ambiguous, loosely-defined business problems, decide which ones are worth solving, formulate them mathematically, and ship optimization systems that run Valmo's network every day. Workstreams span network design, routing and scheduling across the first, middle and last mile.
What you will do
Own a workstream end to end: decide which problems are worth solving, set and drive its roadmap, and be accountable for the movement of its metrics.
Formulate complex, loosely defined business problems in network design, routing, scheduling, allocation and capacity planning as LP, MILP, MINLP or constraint programming models, with explicit assumptions.
Solve them at scale with decomposition (Benders, Lagrangian relaxation, column generation, branch-and-price), heuristics and metaheuristics (large neighbourhood search, ALNS, tabu search) and constraint programming, within real planning-time budgets.
Handle uncertainty with stochastic, robust and chance-constrained optimisation, and test decisions through simulation before they reach the ground.
Combine OR with ML and DL, such as demand and volume forecasting, ETA prediction and learned heuristics, wherever it improves the decision.
Write production-grade code and work with engineering to ship solvers and models into systems that run every day, with testing, monitoring and reproducible pipelines. Use AI coding agents as part of how you build, and verify the formulations, code and results they produce.
Mentor DS1s and DS2s on formulation, modelling craft and code quality; align the roadmap with product, engineering, operations and business; and explain trade-offs to senior leaders in decision terms.
Publish your work at A* conferences and journals, as part of a team with a highly innovative work culture.
What you will need
Bachelor's, Master's or PhD in Operations Research, Industrial Engineering, Applied Mathematics, Statistics, Computer Science or a related quantitative field.
5–8 years of experience (or 3+ years with a PhD) applying operations research to real-world problems, with optimisation models deployed in production.
Deep expertise in mathematical optimisation (LP, MILP, network flows, integer and combinatorial optimisation), with a proven ability to formulate ambiguous problems, design heuristics, and break complex problems into simpler sub-problems.
Experience with large-scale solution methods (decomposition, column generation, heuristics and metaheuristics, constraint programming) and with optimisation under uncertainty and simulation.
Hands-on experience with solvers and modelling tools such as Gurobi, CPLEX, OR-Tools (including CP-SAT), HiGHS, Pyomo or PuLP.
Strong programming: expert Python and SQL, PySpark or similar for large datasets, and solid software engineering (object-oriented design, Git, testing, code review). Fluency with AI coding agents, with the judgement to check their output.
Working knowledge of ML and DL (forecasting, regression, tree-based models, neural networks) and of how to combine them with optimisation.
Experience leading a workstream and mentoring junior data scientists, and the ability to explain trade-offs and limitations to senior business and operations leaders.
Good to Have:
Domain experience in logistics, e-commerce, transportation or supply chain: network design, line haul, vehicle routing (including time windows) or last-mile delivery.
Experience with deep learning based forecasting, learning-augmented optimisation (ML for combinatorial optimisation, reinforcement learning) or GPU-accelerated optimisation.
Performance-critical programming in C++, Java or Julia, and familiarity with cloud platforms such as Google Cloud.
Publications at OR/ML venues (e.g. INFORMS, Transportation Science, Operations Research, NeurIPS, ICML, KDD) or open-source contributions.
