Skip to content

Precious Olusola

AI Software Engineer

I build smart software that solves real problems.

I build AI products end to end: the model, and everything around it that makes it dependable, from the API and background workers to the web and mobile apps. I want to be able to take any idea and turn it into a working, intelligent product, on a phone, on the web, on a desktop or on a piece of hardware.

Open to roles and collaborations · Lagos, Nigeria

AI systems

Turns a photo of a Nigerian bank or fintech receipt (OPay, PalmPay, Kuda, Moniepoint, GTBank and more) into structured transaction data. Open source on npm, about 1,000 downloads.

  1. Receipt image
  2. Vision model on Groq, temperature 0, JSON mode
  3. Normalise enums, coerce numbers
  4. Typed TransactionRecord
  • An unreadable field comes back as null, never a misleading 0.
  • If parsing fails, the raw model output is returned so the caller can debug or retry.
  • Next: An evaluation set of real receipts to measure field-level accuracy per bank.

TypeScript · Groq · Qwen 3.6

Output shape (sample values)
{
  "amount": 25000,
  "currency": "NGN",
  "platform": "opay",
  "transaction_type": "transfer",
  "status": "successful",
  "recipient_account": "0123456789",
  "reference": "250114010100xxxx",
  "transaction_date": "2025-01-14T10:42:00+01:00",
  "narration": null
}

Ask a question about the Bible in plain language and get an answer with chapter-and-verse references. Runs entirely on Cloudflare's edge.

  1. Question
  2. Recent conversation from KV
  3. Llama 3.3 70B on Workers AI, citation-first prompt
  4. Answer with references highlighted
  • Keeps conversation context per session, trimmed to the recent turns that matter.
  • Next: Retrieve verses from the Bible text itself, so every citation can be checked instead of trusted.

Cloudflare Workers AI · KV · React

Picsort

Sorts a large photo collection by the people in it, so everyone can find every photo they appear in. Co-built.

  1. Photos from Google Drive, ZIP or upload
  2. InsightFace face embeddings
  3. HDBSCAN clusters faces into people
  4. FAISS index finds every photo of a person
  • HDBSCAN needs no fixed number of people up front: it finds the groups in the data.

Python · InsightFace · HDBSCAN · FAISS

Where I'm going

  • AI that holds up in production

    Agents, retrieval and structured extraction, plus the evaluation work that shows whether an AI feature actually works.

  • Beyond the screen

    On-device models, embedded systems and hardware, so smart software can live where the problem is.

  • Products people use

    Whatever the platform, the measure is the same: real users and real problems solved.

I haven't built the thing that defines this yet. I'm building toward it in the open.

Product work

  • Fintech

    BCDees

    Helps Nigerians in the diaspora exchange naira for foreign currency safely. The naira is locked in escrow, the foreign currency moves outside the app, and the naira is released only when the user confirms it arrived. BCDees never holds foreign currency itself.

    Node.js · Prisma · PostgreSQL · BullMQ · Redis

  • Proptech

    Hivark

    Lets Nigerian tenants save toward yearly rent monthly, and gives landlords and caretakers one place to run their properties. Never holds funds itself.

    Expo · React Native · Express · PostgreSQL

  • Mobility

    Real-time GPS tracking and NFC tap-to-pay for FUTA's campus shuttle, keke and cab. Started as my final-year project.

    React · TypeScript · GPS · NFC

  • Education

    Learn probability as a game: twelve modules, streaks and leaderboards for secondary school students.

    Next.js · MDX · KaTeX

Experience

  • 2023 – Present

    Full-Stack Developer · BCDEEs

    Built the backend: naira escrow and release, admin tools, and BullMQ workers for transactions, KYC and reconciliation. Integrated Paystack and Veriff.

  • 2022 – Present

    ICT Coordinator & Technical Lead · RCF FUTA

    Lead a small team running the digital systems of a 1,000+ member fellowship: member management, yearbook, library and internal email.

  • 2024

    Full-Stack & AI Integration Engineer · Picsort

    Co-built the face-recognition pipeline and the product around it: Google Drive import, Paystack payments, and faster inference.

B.Tech, Software Engineering · Federal University of Technology, Akure

How I work

  1. Find the root cause before fixing anything, and prove the fix works.

  2. Measure AI, don't just demo it: know how often it's right, and what happens when it's wrong.

  3. Design for failure: retries, clean shutdowns, and no surprises where money moves.