Junior ML / Algorithms Engineer — Contineu

Bangalore · On-site · 0–2 years · Immediate start

---

The work

We turn 360° walkthroughs of active construction sites into accurate 3D reconstructions, then run geometric and semantic reasoning over them — comparing what's actually been built against the design model, finding defects, tracking what changed. The data breaks in ways benchmarks never do: motion blur, bad lighting, heavy occlusion, half-built structures that match no clean prior.

You'll work directly with Kanao (CTO) on the core algorithmic problems underneath this. The scope is deliberately wide and moves as the research does. Recent and near-term problems look like:

• Reconstruction from panoramic (e.g. INSV) video — SfM/MVS, neural reconstruction (NeRF, Gaussian splatting), depth estimation, and making any of it survive real capture conditions
• Registration — aligning reconstructions to each other and to the design model across time
• Algorithms over IFC / BIM — parsing, querying, and reasoning about building geometry and the relationships encoded in a building model
• Segmentation and detection on 2D and 3D construction data
• Whatever the highest-leverage problem is that quarter — the role follows the research, not a fixed ticket queue

This is not a call-the-model-and-ship-the-wrapper role. You're expected to understand what's happening underneath, and to invent when the off-the-shelf approach falls over — which, on this data, is often.

---

What we're looking for

• 0–2 years, with one thing you've built that proves depth — a project, paper, competition (Kaggle, ICPC, research), or portfolio. We care more about how deep you went on one thing than how many things you've touched.
• Strong programming fundamentals — data structures, algorithms, complexity. You can reason about why your code is correct and what it costs.
• Strong ML Fundamentals. You can explain how the methods you've used actually work.
• Bias to action. Fast iteration, low ego, comfortable being wrong quickly and moving on.
• Bonus: hands-on with 3D, computer vision, or geometry — point clouds, multi-view geometry, camera models, SLAM, registration, anything in this space.

We optimize for raw ability and fundamentals over specific-tool experience. If you've never touched IFC or Gaussian splatting but you learn fast and reason cleanly, you're who we want.

---

What it's actually like here

Small team, direct line to the CTO, no layers. You own problems end to end and watch them hit production fast. Cloud and GPU infra is there — use it well. Real data, real users, real constraints. Not a research sandbox.

---

How to apply

Send us one paragraph about something you find genuinely fascinating — a topic, field, or idea. It doesn't have to be ML or even technical. We want to see how you think when you're chasing something you care about: how deep you go, how you dig in, what you're doing in that area right now.

Attach your resume and send both through the application link.

---

Process

1. Resume screen
2. Intro call
3. Technical round 1 — work through real problems we actually face
4. Technical round 2 — deeper problem-solving and fundamentals
5. Culture fit