Victor Omolaiye

AI Solutions Engineer helping SMEs & NGOs turn AI into practical, dependable systems: RAG, LLM workflows, privacy-first AI. Free AI Readiness Checklist ↓

[Victor Omolaiye @iamvictorok]

1/ I built IntelliPlug, a support assistant that answers customers from each company's own documents, with every company's data kept separate.

Here's how it works, and what moved the needle. 🧵

[Victor Omolaiye @iamvictorok]

Teachers spend evenings writing lesson plans and assessments.

Teacher Co-Pilot drafts curriculum-aligned lesson plans and assessment materials for secondary schools, and a second AI step checks each draft before the teacher sees it.
https://t.co/2ONqBRc3LF

[Victor Omolaiye @iamvictorok]

@nishant_zk @SmartJellyAI I'd like to help. I build RAG and knowledge assistants: retrieval, evals, answers grounded in sources, which is close to what Jelly's memory does. Happy to start on a defined piece and show work. Sending you a DM.

[Victor Omolaiye @iamvictorok]

@pablothethinker This has been really impressive

[Victor Omolaiye @iamvictorok]

5. Which cost am I choosing to carry: hours, hosting, upkeep?

Engineering is applied philosophy. Full write-up:
https://t.co/wGft0b2TqW

[Victor Omolaiye @iamvictorok]

A system can pass every test and still fail in its first week with real people: bad wifi, forgotten passwords, zero patience, sometimes bad intentions.

The spec describes an ideal world. Nobody lives there. My checks:

[Victor Omolaiye @iamvictorok]

1. Why does this need to exist at all? Delete before you build.
2. Does it survive someone who never reads the docs?
3. Is a retry safe, or does it corrupt data?
4. If the complexity can't go, can I absorb it instead of handing it to the user?

[Victor Omolaiye @iamvictorok]

4. You get a ranked list and a report to share with the hiring manager.

It doesn't replace the recruiter. It gives them a consistent first pass, so time goes on conversations.

Case study: https://t.co/xqNhZFk97s

[Victor Omolaiye @iamvictorok]

A job ad goes up. By Friday the shared inbox is full of CVs. For small HR teams, the bottleneck isn't finding people. It's reading them.

I built CV Ranker for that. How it works:

[Victor Omolaiye @iamvictorok]

1. It reads the CVs (PDFs too) and the job description.
2. It compares them by meaning, not only keywords.
3. A language model adds a reasoned view per candidate.

[Victor Omolaiye @iamvictorok]

@adcock_brett Its great, easily the most impressive LLM product I have used since chatgpt first launched. I am left with 30+% of the monthly h3 allocation already and its been amazing

[Victor Omolaiye @iamvictorok]

It runs as a stateless service behind a REST API, so other systems can ask it for a timetable.

Case study: https://t.co/kqns4NtR0N

[Victor Omolaiye @iamvictorok]

Every term, someone at a school loses days to a puzzle: one teacher in two rooms, the lab booked twice, fix one clash and two more appear.

I built a timetable scheduler for it. What worked, in 3 posts:

[Victor Omolaiye @iamvictorok]

2. Then mutate where the clashes actually are. Conflict-directed mutation converged to schedules with zero clashes.

Lesson: in scheduling, a good starting point beats brute force.

[Tech_baby @Tech_babby]

Learn to create content.

Learn Problem solving.

Learn to sell.

Learn AI Leverage.

Learn to build with AI.

Learn Distribution.

These are the high end skills I recommend everyone learns in 2026. Did I miss anything? 👨🏾‍💻👀

[Victor Omolaiye @iamvictorok]

A model in a notebook has no users, no uptime, and no way for another system to call it.

Value starts when it becomes a service: state, auth, streaming, background jobs.

That's backend work, and AI engineers need it.

https://t.co/flwE7QDYKb

[Victor Omolaiye @iamvictorok]

Minutes from a sensitive meeting, without uploading the recording anywhere?

I built Sovereign Minute Maker for that. Transcription and personal-detail redaction run on your device. Only cleaned text reaches the LLM.

https://t.co/OzuMqGZyNc

[Victor Omolaiye @iamvictorok]

@de_henne Nice result. Scope the MCP plugin to only the skills it needs and log the tool calls, so you can spot when a skill stops improving the click. The guardrails are what let automation compound.

[Victor Omolaiye @iamvictorok]

@mardehaym Masking alone still leaks geography, so testing outcomes per release matters more than the prompt. What moves these builds: map sources before the model touches a file, then eval historical files to gate each deploy.