Tochukwu Onwukwe
AI automation engineer and AI trainer. I build API-driven automations and AI agents that connect the tools a team already uses, and I evaluate AI systems for a living, so I know how they fail.
Lagos, Nigeria (WAT) · Remote · Available full-time
Featured Projects
Built for client work. Code and screenshots are confidential, so each case study covers the approach.
Sales Pipeline Automation
Captures leads from Tally forms, Google Forms, and email, enriches them through APIs, logs them in HubSpot and Google Sheets, and alerts the sales team in Slack. Cut manual lead handling and got new leads to the team right away.
n8n · Claude · HubSpot API · Google Sheets · Slack API
Read the case study →Incoming Data Classifier and Router
Receives data through a webhook or API, classifies it with an LLM, and routes it to the right department. Batching keeps it reliable on large inputs. Reduced handling time by removing manual sorting.
n8n · Gemini / Claude · OpenAI · Slack API · REST
Read the case study →Company RAG Assistant
A chat assistant that answers team questions from company documents, customer information, and APIs, with a knowledge base refreshed regularly. Saved time searching for information and cut repeat questions.
n8n · Claude / Gemini · n8n Vector Store · REST
Read the case study →More Automation Work
- Order processing pipeline. Webhook-triggered, with API-key authentication, required-field validation, and a modular sub-workflow that checks for duplicate orders and updates status automatically.
- Customer service agent. Gemini and OpenAI agent with tool calling for order status, customer, and product lookups, plus persistent memory.
- Sales data pipeline. Branching workflow that aggregates and validates sales data by region, then exports CSV and emails reports to stakeholders.
- RAG agents. One for open research with SerpAPI search grounding, one for document-grounded Q&A.
- Reliability. Retry and error-handling configurations tested through dedicated failure workflows, and a feedback pipeline debugged end to end after recurring HTTP 422 errors.
- DataBridge Partners. Two-workflow recruitment referral automation: an intake and tagging workflow paired with a scheduled email workflow.
What I Build
- Lead and sales pipeline automation
- LLM-based classification and routing of incoming data
- RAG assistants over company documents and APIs
- Slack, HubSpot, and API integrations
- Error handling, retries, and failure testing for automations
AI Training and Evaluation
I evaluate and improve AI outputs across several platforms, using rubric-based methods and RLHF practices. This shapes how I build: I design for edge cases and add review steps where accuracy matters.
micro1 · AI Trainer / Data Annotator
- Earned micro1's Image Annotator / AI Trainer certification.
- Scored multimodal model responses (image, video, text) for factual accuracy, reasoning quality, and completeness against rubrics.
- Documented hallucinations and inconsistencies with evidence-backed explanations instead of pass/fail labels, which reviewers cited when approving submissions.
AfterQuery · Web Research Analyst / Data Annotator
- Built multi-hop research questions by working backward from a verified answer.
- Validated each question by confirming leading LLMs failed it and five separate Google searches surfaced nothing on page one.
- Reviewed AI responses for bias and toxicity against platform guidelines.
AtlasCapture · AI Trainer / Data Annotator
- Spotted that generic "hold" labels were double-counting objects already captured by another action, and proposed a conditional rule that became the standing guideline.
- Applied the rule across large volumes of video segments, eliminating double and over-labeling.
Test.io · QA / Tester
- Assessed bug severity across a 28-scenario dataset using impact and reproducibility criteria.
- Wrote structured bug reports with a strict What/Where/When title format, clear reproduction steps, and separated expected versus actual results. The format was adopted as the project standard.
Data Quality Dispute Resolution
Fact-checked flagged dataset responses independently and caught two errors other reviewers missed: asbestos described as synthetic, and a song credited to the wrong artist.
Skills
Education and Languages
B.Eng. in Agricultural Engineering, Federal University of Technology Owerri (FUTO), 2019 – 2023
English (fluent) · Japanese (intermediate)
Contact
Open to client projects and full-time remote roles.