Data Scientist, and Quant Professional

Oslo, Norway
Joined August 2019
4/10 Practical Paths into Data & AI, build pathway: pick one bounded AI workflow and create an eval set before adding features. Include normal, edge and failure cases, plus a threshold that blocks release. Google guide: developers.google.com/machin… #AIEngineering
11
3/10 Practical Paths into Data & AI: build pipelines that fail visibly. Check schema, freshness and duplicate IDs. Quarantine bad rows, then prove reruns are safe. Free self-paced Data Engineering Zoomcamp, not a vacancy: github.com/DataTalksClub/dat… #DataEngineering
1
11
2/10 Practical Paths into Data & AI: analytics engineering. Define paid_orders, then test unique IDs, missing values and allowed statuses. The proof is a trustworthy metric, not a prettier chart. dbt data tests (build guide, not a vacancy): docs.getdbt.com/docs/build/d…
15
1/10 Practical Paths into Data & AI: choose a lane. Analytics: one tested metric. Data engineering: one resilient pipeline. Applied AI: one evaluated feature. Start with a free Microsoft Learn pathway, not a job listing: learn.microsoft.com/en-us/tr… #DataCareers
7
MIT’s HardFlow lets generative AI explore during sampling, then enforces hard constraints only on the final output. In tests across robotics, maze navigation and image editing, it achieved 100% constraint satisfaction. news.mit.edu/2026/new-method… #AI #Robotics
1
15
NASA and IBM open-sourced a lunar AI model built from 30+ aligned data layers across 4 missions. It maps craters, volcanic features and possible ice, cutting lunar-ice RMSE by up to 22% vs SwinV2-B. research.ibm.com/blog/nasa-i… #AI #SpaceTech
1
78
Researchers robotically scanned 960 conditions of the 1891 Biginelli reaction and found an unexpected bicyclic product. A mechanism-aware AI then proposed a pathway supported by follow-up experiments. doi.org/10.1038/s44160-026-0… #AI #Chemistry #Robotics
28
Most medical AI models hunt for one disease. Alibaba DAMO’s new RADAR reads one contrast CT across 18 abdominal organs and screens 146 findings. In Science, it averaged AUC 0.913 across 39,160 exams; code is open. github.com/alibaba-damo-acad… #AI #MedAI
39
GiveDirectly is hiring a remote Senior Software Engineer to build cash-transfer systems. Nigeria approved; local work authorization required. Needs 4+ years and backend/API/cloud experience. Deadline not stated. Apply: job-boards.greenhouse.io/giv…
98
AI agents are inventing slang. In Emergence World Season 2, agents formed shared terms without being told to, and their language grew harder for humans to interpret. The twist: readable logs may not mean understandable behavior. world.emergence.ai/ #AI #AIAgents
1
40
A 44.1 g drone can navigate by touch in complete darkness. Two whiskers feed a neural depth model + Kalman filter, all running onboard with just 34 KB of RAM. Published Sept 18 in Nature Communications. nature.com/articles/s41467-0… #AI #Robotics
19
Stanford’s “Virtual Biotech” used 37,000 AI agents to analyze 55,984 clinical trials in under a week, then proposed a lung-cancer drug strategy. The agents were structured like a real biotech, with a virtual CSO and specialist teams. virtualbiotech.ai/ #AI
21
UX designers in Nigeria and across Africa: Canonical is hiring a remote EMEA UX Designer, Design systems, to build reusable UI. Needs a design-systems portfolio + Figma/Storybook. Travel up to 4x/year. Deadline not stated. canonical.com/careers/704302… #UXJobs
128
₦5m at Dangote Refinery's ₦525 IPO price buys 9,520 shares. At hypothetical 3%, 5% or 8% dividend yields: ~₦150k, ₦250k or ₦400k gross/year. Dividends aren't guaranteed. The real debate: what would justify the share price multiplying? @DangoteGroup #DangoteIPO #NGX
28
Oilfield computer vision turns periodic HSE checks into continuous hazard detection. Before alerts influence action, test missed hazards across low light, occlusion, PPE variation and camera failure, with an explicit abstain path. Source: jpt.spe.org/ai-eyes-on-the-r…
16
Rovuma LNG plans 18 subsea wells, ~50 km of flowlines and 12 eLNG modules. Optimising each asset separately can miss the system bottleneck. AI can test coupled well-to-plant scenarios faster, while physics and safety define the feasible envelope. Source: corporate.exxonmobil.com/loc…
1
42
EIA forecasts U.S. crude output at 13.8 mb/d in 2026 and 14.3 in 2027. AI can test more paths than point forecasting, but capital decisions need probabilities across decline, completions and project timing, not one precise headline. Source/visual: eia.gov/outlooks/steo/
8
Drilling AI needs evidence grounding, not fluency. A Volve system cross-references measurements and daily reports instead of manual lookup. Operational answers must preserve well, time and source lineage. Source: arxiv.org/abs/2605.00060 Photo: BSEE, public domain.
5
Rig computer vision should be judged by useful interventions, not detection accuracy. IOGP recorded 946 lost-time injuries in 2024. Continuous monitoring challenges periodic observation, but false alarms can train crews to ignore it. Source: JPT/SPE. Photo: Atmadeep Das/Pexels.
16