Open to AI Engineer roles

I build AI systems that do the work, not just demo well.

I'm Joshua, an AI engineer and founder of VolyxAI. I turn models into practical tools with clear controls, reliable backends, and humans in the loop where judgment matters.

Example of a controlled AI workflow that validates input and routes consequential decisions for human review.
~/ controlled-agent running
$ run workflow --safe
  1. 01Input receivedConversation + required fieldsready
  2. 02Rules checkedValidation before model callpass
  3. 03Decision routedConsequence detectedreview
  4. 04Context preservedOperator receives the full trailqueued

Models into products

·

Backends that hold up

·

Automations with boundaries

·

Evaluation before claims

Selected work

Four working systems. One archived experiment.

Multimodal AI, real-time state, security scanning, and multi-chain API aggregation lead the portfolio. The forecasting experiment remains visible for its evaluation lessons, not as a product claim.

02 / Real-time systemsJavaScript

Noughtline

A React and Express multiplayer game with signed guest sessions, server-validated moves, reconnect handling, an append-only transaction ledger, and one-time crediting for verified Paystack references.

03 / API integrationPython

VirusTotal Telegram Bot

An async Telegram bot for scanning files, URLs, and hashes through VirusTotal v3. It hashes locally before upload, follows the large-file upload flow, backs off on rate limits, and caches reports for interactive details.

04 / API aggregation prototypePython

Solana & Ethereum Wallet Analyzer

An async Telegram bot that validates Solana and Ethereum addresses, combines Solana RPC, Etherscan, CoinGecko, and DexScreener data, caches requests for five minutes, and paginates Solana token holdings in chat.

05 / Archived ML experimentPython

Football Predictor

An archived forecasting experiment combining calibrated XGBoost and Poisson score probabilities. Its real-data evaluation demonstrated signal over naive baselines but not practical value over bookmaker probabilities.

  • Rolling-origin evaluation across 1,140 Premier League test matches
  • 53.77% accuracy versus a 56.70% bookmaker benchmark; not a betting product

Founder build · VolyxAI

Useful automation needs a boundary.

At VolyxAI, I'm working on controlled operations workflows: intake, validation, approval, and handoff. The model can help reason about messy input. Deterministic rules and clear ownership decide what happens next.

Role
Founder · AI and automation engineering
Status
Early-stage · workflow shown is illustrative
Decision
Consequential actions wait for human approval
Visit VolyxAI
Controlled workflow: receive an input, validate it, request human review, then hand off the approved action with its audit trail.
01Receivevoice · message · webhook
02Validateschema · rules · context
03Reviewperson owns the decision
04Handoffaction + full audit trail
ABOUT / 05

About

An engineer who follows the system past the model call.

I work across AI integration, backend engineering, and automation. That means tracing failures through webhooks, databases, queues, auth, retries, and the awkward edge cases between them.

I care about honest evaluation, readable code, and systems another person can operate without calling the original builder every time something breaks.

Model work

Integration · retrieval · evaluation · structured outputs

Backend systems

FastAPI · Express · Socket.IO · REST · webhooks

Languages

Python · JavaScript · TypeScript · SQL

Automation

n8n · approval gates · retries · idempotency

Start a conversation

Hiring an AI Engineer who can build the system around the model?