Kelvin Kirimi

Key Skills

Software

AWS SageMaker

Argilla

CVAT

Dataloop

Diffgram

Doccano

Encord

Kili Technology

Labelbox

Label Studio

Prodigy

Roboflow

SuperAnnotate

Supervisely

Tagtog

V7 Labs

Appen

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Image

Text

Video

Top Task Types

Audio Recording

Bounding Box

Classification

Entity Ner Classification

Segmentation

Freelancer Overview

Data Annotation Specialist with 3+ years delivering 99.7% QA across computer-vision, NLP, and audio/video datasets. 150K+ CV frames (bounding boxes, polygons/masks, keypoints, tracking) for autonomous-driving, retail product/shelf, and defect detection. 20K+ text items (NER, sentiment, intent, taxonomy, content safety) plus hours of A/V transcription with speaker ID and event tagging. Strengths: rapid spec ramp-up, edge-case discovery/escalation, guideline drafting/refinement, gold-set creation, inter-annotator agreement (IAA) checks, audit-ready QA. Convert ambiguous tickets into clear steps and measurable metrics; keep rework <0.5%. Toolbox: CVAT, Labelbox, Label Studio, SuperAnnotate, Prodigy, V7, Encord, Argilla, AWS SageMaker GT; Jira/Confluence, Sheets/Excel QA dashboards. Comfortable with model-in-the-loop feedback and data-drift notes. Collaboration: mentored 5+ juniors (onboarding −30%), deliver 100% on time across time zones, NDA-friendly, remote-first. Immediate availability.

Labeling Experience

Lead Image & Video Annotator — Autonomous Driving

CVATVideoBounding BoxPoint Key Point

Annotated 150,000+ frames with boxes, masks, tracking, lane polylines, and keypoints in CVAT. Reached 99.7% QA using gold sets + IAA checks and documented 300+ edge cases (night/rain/dusk/occlusion). Guideline iterations with engineers cut model false-negatives by 12%; delivered audit-ready exports and concise QA dashboards.

Annotated 150,000+ frames with boxes, masks, tracking, lane polylines, and keypoints in CVAT. Reached 99.7% QA using gold sets + IAA checks and documented 300+ edge cases (night/rain/dusk/occlusion). Guideline iterations with engineers cut model false-negatives by 12%; delivered audit-ready exports and concise QA dashboards.

2023

Geospatial Satellite Imagery Classifier — Land Cover Mapping

LabelboxGeospatial Tiled ImageryClassificationMapping

Labeled 5,000+ km² of multispectral satellite imagery into land‐cover classes (forest, water, urban, agriculture) for an environmental analytics firm. Created custom polygon masks and ensured edge‐to‐edge tile consistency. Conducted QA spot‐checks on 10% of tiles, achieving 96% accuracy against ground truth. Supplied georeferenced shapefiles compatible with client GIS workflows.

Labeled 5,000+ km² of multispectral satellite imagery into land‐cover classes (forest, water, urban, agriculture) for an environmental analytics firm. Created custom polygon masks and ensured edge‐to‐edge tile consistency. Conducted QA spot‐checks on 10% of tiles, achieving 96% accuracy against ground truth. Supplied georeferenced shapefiles compatible with client GIS workflows.

2023 - 2023

Lead Image & Video Annotator — Autonomous Driving

CVATVideoBounding Box

Annotated 100+ hours of dash-cam video for a major self-driving dataset. Drew precise bounding boxes around vehicles, pedestrians, cyclists, and traffic signs, and created per-frame semantic segmentation masks for drivable areas and lane markings. Collaborated with QA lead to define annotation guidelines, ran weekly calibration sessions, and maintained ≥ 99.5% accuracy against ground truth.

Annotated 100+ hours of dash-cam video for a major self-driving dataset. Drew precise bounding boxes around vehicles, pedestrians, cyclists, and traffic signs, and created per-frame semantic segmentation masks for drivable areas and lane markings. Collaborated with QA lead to define annotation guidelines, ran weekly calibration sessions, and maintained ≥ 99.5% accuracy against ground truth.

2023 - 2023

Senior Audio Transcription & Classification — Call Center Speech

Aws SagemakerAudioClassificationEmotion Recognition

Transcribed and classified 50+ hours of multilingual (EN/ES) call recordings. Labeled intent tags (billing, support, cancellation) and sentiment scores, achieving 98% transcription accuracy and 95% classification recall. Built speaker-diarization segments, trained junior annotators on guidelines, and delivered CSV exports with timestamps for downstream analytics.

Transcribed and classified 50+ hours of multilingual (EN/ES) call recordings. Labeled intent tags (billing, support, cancellation) and sentiment scores, achieving 98% transcription accuracy and 95% classification recall. Built speaker-diarization segments, trained junior annotators on guidelines, and delivered CSV exports with timestamps for downstream analytics.

2023 - 2023

Senior 3D Point-Cloud Cuboid Annotator — Robotics & LiDAR

Scale AI3D SensorCuboidEvaluation Rating

Annotated 200 million+ LiDAR points into 3D cuboids for obstacle and object detection in autonomous robotics research. Defined and enforced strict annotation guidelines for vehicle, pedestrian, and static‐structure classes. Ran weekly inter‐annotator agreement checks, achieving 98% IAA, and delivered gold‐standard datasets that improved model precision by 15%.

Annotated 200 million+ LiDAR points into 3D cuboids for obstacle and object detection in autonomous robotics research. Defined and enforced strict annotation guidelines for vehicle, pedestrian, and static‐structure classes. Ran weekly inter‐annotator agreement checks, achieving 98% IAA, and delivered gold‐standard datasets that improved model precision by 15%.

2022 - 2022

Show more

Education

Work History