Teaching AI models
to sound more human.
Six years into training and evaluating the systems that train language models - rating responses, rewriting weak ones, and building the QA processes that keep a 90-client annotation practice at 99%+ accuracy.
- English (Fluent) · Yoruba (Native)
Impact at a glance
An AI Trainer and Entrepreneur
I'm Mercy, the COO and co-founder of Sellyt, an AI-powered platform built to transform how vendors sell online. Millions of talented vendors are hustling daily across WhatsApp, Facebook, and Instagram with no proper tools to help them grow, so we built one.
My role is to keep everything moving, from operations and team building to strategy and community. Africa's best days are ahead, and I'm growing up every day to help build that future.
How I show up every day
Operations
Running the day-to-day of Sellyt and keeping all moving parts in sync.
Team building
Recruiting, managing, and growing the Sellyt team from the ground up.
Strategy
Planning and executing Sellyt's expansion across West Africa.
Community
Growing and nurturing the Sellyt vendor community across the platform.
Partnerships
Building relationships with investors, partners, and stakeholders.
Innovation
Driving product thinking and new ideas that keep Sellyt ahead.
Milestones so far
- Co-founded Sellyt and led it from idea to a live platform serving vendors at scale.
- Successfully onboarded 1,000+ beta vendors and built an active vendor community.
- Structured and led the hiring process for the Sellyt team from scratch.
"I Don't Just Use AI - I Teach It."
"Before building Sellyt, and alongside running it, I work as a professional AI Trainer and Annotation QA specialist, helping shape the way artificial intelligence thinks, responds, and understands the world. It's one of the most fascinating and important jobs of our generation, and I'm proud to be part of it."
Entrepreneurship training
2026 MTN Foundation–BOI Y'ellopreneur 3.0 Entrepreneurship Skills Training
Facilitated by the Enterprise Development Centre, Pan-Atlantic University, in partnership with MTN Foundation and Bank of Industry.
View certificate (PDF)Client's feedback
"Mercy did a great job!"
What she brings to a project
AI Tutoring & Training
LLM Training Support
Multimodal Data
Annotation Tools
Writing & Communication
Quality & Leadership
Six years of labels, mostly under NDA - a few she can show
Only a fraction of Mercy's project history is shareable; most client work is covered by NDA. These samples span image, text, document and audio annotation.
Labeled pedestrians, vehicles, trees and road signs at pixel level across high-resolution street scenes, with layered peer and supervisor QA to keep class boundaries consistent frame to frame.
Classified customer-service replies by intent, tone and resolution status, tagging whether the brand response matched the user's frustration and if the issue was resolved.
Reviewed claim–evidence pairs on climate topics, researching source context before marking each sentence as supporting, refuting, or insufficient - logic that carries straight over to RLHF fact-checking.
Tagged names, dates, line items and totals across handwritten and printed invoices, producing consistent, audit-ready exports for downstream model training.
Tagged job postings by section - role, requirements, benefits, salary - against a fixed category list, keeping labels consistent across hundreds of listings.
Segmented and classified call-audio waveforms by event type - voicemail, human speech, phone tree, ringing - using a controlled vocabulary to keep boundaries clean across long recordings.
A record of QA-led project work
- Runs routine quality checks and tests across image, text and video labeling tasks
- Builds and implements QA procedures, documentation and policy
- Identifies and resolves workflow and production issues across teams
- Writes training materials and operating manuals, and documents every audit
- Rated text outputs on tone, accuracy and relevance to teach models safer, more helpful responses
- Wrote curated examples used directly in RLHF and SFT training
- Worked with engineers to refine annotation tools and workflows
- Reviews 1,500 SKUs a week, extracting attributes and assigning HS codes at >97% first-pass accuracy
- Flags compliance risk, cutting customs-hold incidents by 18%
- Maintains the annotation logs and SOPs junior annotators now use daily
- Held 99% annotation accuracy training an AI model on audio data
- Onboarded new hires on audio annotation practice
- Turned model-testing results into reports the AI team used to improve performance
- Shared weekly sprint dataset releases with the AI team
- Delivered multimodal datasets - text, audio, video, image - to 90+ clients worldwide at 99% QA
- Recorded and labeled voice samples for conversational AI, tuned for natural dialogue
- Wrote the guidelines and dialogue scenarios used to train generative AI systems
- Managed projects end to end: scoping, onboarding, scheduling, QA
- Annotated video, image and text data for model training
- QC'd team output and halved rework rates
- Trained junior annotators, lifting both throughput and accuracy