Lazarus Danjuma

Glad to see several persons got selected for the Sovrano AI internship I dropped here few days ago. I will always be at the fore front of sharing opportunities to you all, no matter where you stay or reside. However, you need to be skilled or talented enough to actually get selected during the process of this applications. As a lot of other processes are beyond my control, and more on the team making the decision or hiring.

It is seriously not ideal to meet someone today and start asking them for a job or to recommend you for one. The truth is, nobody will recommend someone they do not know or trust for a job opportunity. The best way to go about it is: build a relationship, prove your personality, prove your competency, and let them believe you can be trusted. Good morning 🙌

Glad to be Selected for the Sovrano AI internship program(Duration: 12 weeks (€500/month). Can’t wait for what’s next and the opportunity to contribute with a team building Europe’s evaluation layer for AI. Glad to see Marcus Castro and Same team that built CareerOS. Bridging careers and students through Sovrano. With the aim to connect students with the AI companies shaping the world, through university partnerships across Europe. If you got the talent, skills or expertise, and you believe you should have a shot at this, come join us at Sovrano, let’s build something great together. PS: Link to the internship program in the comment

Spending my weekend deep in a Segmentation & Labeling SOP for ego-view video annotation, the kind of dataset that teaches humanoid robots how humans actually do physical work. Two jobs: cut the video into clean segments, then label what each hand does. Sounds simple. It is not. A few things that reshaped how I think about annotation quality: 𝟭. 𝗧𝗵𝗲 𝟱-𝗳𝗿𝗮𝗺𝗲 𝘃𝘀 𝗵𝗮𝗹𝗳-𝘀𝗲𝗰𝗼𝗻𝗱 𝗿𝘂𝗹𝗲 Not every cut deserves the same precision. A grab, a let-go, a screw meeting a screwdriver, that's a contact-based boundary, audited to ±5 frames. But a hand just starting to reach toward an object? That's intent-based, and you place it roughly, within half a second. Knowing which cue you're looking at before you touch the timeline saves so much rework. 𝟮. "𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗼𝘂𝘀" 𝗶𝘀𝗻'𝘁 "𝗶𝗻𝗱𝗶𝘃𝗶𝘀𝗶𝗯𝗹𝗲" My first instinct with something like mopping or scrubbing was to treat it as one long action. Wrong. Cyclic tasks get split into 10-second-or-less pieces, no exceptions, the "long indivisible task" rule only applies when there's a genuine definite end that truly can't be split (think: fitting a plastic liner). Repetition ≠ a free pass. 𝟯. 𝗜𝗱𝗹𝗲 𝘁𝗶𝗺𝗲 𝗵𝗮𝘀 𝗮 𝗰𝗲𝗶𝗹𝗶𝗻𝗴 Neither hand doing useful work? It gets its own "idle" segment, capped at 5 seconds. Cross that, and the entire clip gets rejected before delivery. A small detail that completely changes how you triage a video before you even start cutting. 𝟰. 𝗟𝗮𝗯𝗲𝗹𝘀 𝗮𝗿𝗲 𝗻𝗼𝘁 𝗻𝗮𝗿𝗿𝗮𝘁𝗶𝗼𝗻 Four rules, non-negotiable: imperative verb ("Slice the bread," not "Plugging cable in" or "Begin plugging"), self-contained (no "continue," no "finish"), hand always named, physical action only, never intention. "Inspect the bottle" fails instantly because inspecting is a thought, not a motion. "Pick up the bottle with the left hand" is what the model can actually see. 𝟱. 𝗧𝗵𝗲 𝘃𝗮𝗴𝘂𝗲-𝘃𝗲𝗿𝗯 𝘁𝗿𝗮𝗽 Manipulate. Adjust. Handle. Move. These feel descriptive but they're almost always banned or conditional. "Move" only survives if there's an endpoint ("move iron to table"). "Adjust" only survives as a minor position fix with no clearer verb available. Precision in word choice is the whole job. The edge case that stuck with me most: two hands working at once, doing different things, in sequence, you write both actions, in the correct time order, because the model has no way to infer sequence from a single frozen label. Annotation is basically teaching a machine to see causality it can't otherwise perceive. This is the part of AI work nobody posts about, not the model outputs, the invisible scaffolding underneath them. Every humanoid robot demo you'll see in the next few years is downstream of someone deciding, frame by frame, whether a reach counts as a cut. Grateful to be one of the people doing that deciding. #AIAnnotation #DataAnnotation #HumanoidRobotics #RemoteMondays #AITraining #EgoViewData #GroundTruth

Spending my weekend deep in a Segmentation & Labeling SOP for ego-view video annotation, the kind of dataset that teaches humanoid robots how humans actually do physical work. Two jobs: cut the video into clean segments, then label what each hand does. Sounds simple. It is not. A few things that reshaped how I think about annotation quality: 𝟭. 𝗧𝗵𝗲 𝟱-𝗳𝗿𝗮𝗺𝗲 𝘃𝘀 𝗵𝗮𝗹𝗳-𝘀𝗲𝗰𝗼𝗻𝗱 𝗿𝘂𝗹𝗲 Not every cut deserves the same precision. A grab, a let-go, a screw meeting a screwdriver, that's a contact-based boundary, audited to ±5 frames. But a hand just starting to reach toward an object? That's intent-based, and you place it roughly, within half a second. Knowing which cue you're looking at before you touch the timeline saves so much rework. 𝟮. "𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗼𝘂𝘀" 𝗶𝘀𝗻'𝘁 "𝗶𝗻𝗱𝗶𝘃𝗶𝘀𝗶𝗯𝗹𝗲" My first instinct with something like mopping or scrubbing was to treat it as one long action. Wrong. Cyclic tasks get split into 10-second-or-less pieces, no exceptions, the "long indivisible task" rule only applies when there's a genuine definite end that truly can't be split (think: fitting a plastic liner). Repetition ≠ a free pass. 𝟯. 𝗜𝗱𝗹𝗲 𝘁𝗶𝗺𝗲 𝗵𝗮𝘀 𝗮 𝗰𝗲𝗶𝗹𝗶𝗻𝗴 Neither hand doing useful work? It gets its own "idle" segment, capped at 5 seconds. Cross that, and the entire clip gets rejected before delivery. A small detail that completely changes how you triage a video before you even start cutting. 𝟰. 𝗟𝗮𝗯𝗲𝗹𝘀 𝗮𝗿𝗲 𝗻𝗼𝘁 𝗻𝗮𝗿𝗿𝗮𝘁𝗶𝗼𝗻 Four rules, non-negotiable: imperative verb ("Slice the bread," not "Plugging cable in" or "Begin plugging"), self-contained (no "continue," no "finish"), hand always named, physical action only, never intention. "Inspect the bottle" fails instantly because inspecting is a thought, not a motion. "Pick up the bottle with the left hand" is what the model can actually see. 𝟱. 𝗧𝗵𝗲 𝘃𝗮𝗴𝘂𝗲-𝘃𝗲𝗿𝗯 𝘁𝗿𝗮𝗽 Manipulate. Adjust. Handle. Move. These feel descriptive but they're almost always banned or conditional. "Move" only survives if there's an endpoint ("move iron to table"). "Adjust" only survives as a minor position fix with no clearer verb available. Precision in word choice is the whole job. The edge case that stuck with me most: two hands working at once, doing different things, in sequence, you write both actions, in the correct time order, because the model has no way to infer sequence from a single frozen label. Annotation is basically teaching a machine to see causality it can't otherwise perceive. This is the part of AI work nobody posts about, not the model outputs, the invisible scaffolding underneath them. Every humanoid robot demo you'll see in the next few years is downstream of someone deciding, frame by frame, whether a reach counts as a cut. Grateful to be one of the people doing that deciding. #AIAnnotation #DataAnnotation #HumanoidRobotics #RemoteMondays #AITraining #EgoViewData #GroundTruth

Are you good in Basic C++ & C ++? A company I work for is hiring technical experts in that domain (200 openings so hurry) What you need to know about the application process; Read the job description, at the top you will see the pay rate per hour Then proceed to register or Apply for the job. In the Application form you will put in; ✅Name ✅Last Name ✅Email ✅Phone Number ✅LinkedIn URL ✅Upload CV/resume Before you start the Real interview (DO THIS) The most important part is passing the AI interview with Zara, so you secure the job. Here is how to prepare to pass it on One shot; Use the micro1 perp website, this will enable do a mock or simulation test: ✅micro1.ai/interview-prep ✅Search for the exact or similar role and do the mock interview ✅Once you have gained good confidence, you can now Do the main interview. ➡️Note this interview has to occur with a LAPTOP and not your Mobile Phone (Recommend good internet and Firefox browser) Job 1: Technical Problem Solver / Puzzle Solver Link: https://lnkd.in/eaDQU8wr Job 2: Competitive Programmer Link: https://lnkd.in/esqZDdSm

Micro1 is selecting expert AI Trainers & AI Data Annotation Expert to participate in a customer-facing project focused on advancing AI capabilities. In this role, you'll apply your expertise to help train next-generation AI systems. Hiring Now- AI Data Annotation Expert (100 openings) Apply now: https://lnkd.in/eayT5fkG Hiring Now- AI Trainer (100 openings) Apply Now: https://lnkd.in/eA5ffubU

Feels like there is a surge in hiring for technical experts across all domain. Know anyone? Retweet or tag them. I’m building something massive

If you’re trying to land a remote job before the end of the year, pay attention to these two countries: Australia 🇦🇺 and the UAE 🇦🇪. Australia is booming with e-commerce brands, Shopify stores, digital agencies, SaaS companies, and service businesses hiring remote talent. The UAE is growing rapidly in real estate, fintech, hospitality, luxury brands, and startups that are actively looking for remote professionals. Go to your preferred search engines Search for some of these keywords : Shopify Agency Australia E-commerce Brand Australia SaaS Company Australia Real Estate UAE Fintech Dubai Digital Agency UAE Then connect with their founders, recruiters, or hiring managers on LinkedIn, X or other social media platforms. That’s where many remote opportunities are found before they ever make it to job boards.