A.R.O. emblem

AYODEJI
OGUNLAMI

AI Engineer · Forward Deployed Engineer · Agentic AI · Databricks · Azure

I design and ship production AI systems for enterprise clients: LLM applications, agents, and RAG, taken from scoping to deployment.

Portrait of Ayodeji Ogunlami
5+
Years in data & AI
13
Projects on this page
6
Certifications

Profile

AI engineer and consultant with 5+ years turning business problems into production-grade AI. At Advancing Analytics I design and ship LLM, RAG, and agentic systems on Azure and Databricks for enterprise clients across insurance, legal, and regulated domains, taking solutions from scoping and requirements through deployment and support.

I build AI that teams can trust in production: rigorous evaluation, guardrails on model inputs and outputs, privacy protection, and human review on consequential decisions. Certified ScrumMaster and Azure-certified, with a foundation in geospatial science that keeps me precise about ground truth.

Skills

LLMs & Generative AI
LLM applicationsRAG pipelines Prompt engineeringEvaluation ClaudeGPT-4oGemini
Agentic AI
LangGraphMCP integration Multi-agent systemsn8n automationHuman-in-the-loop
Cloud, Data & Delivery
Azure OpenAIAzure MLAzure AI Search DatabricksMLflowPostgreSQL PythonSQLTypeScript Scrum / AgileGeospatial & GIS

Experience

JAN 2026 – PRESENT · LONDON / UK · REMOTE
Forward Deployed Engineer
Advancing Analytics
  • Embed directly with enterprise clients as a hands-on engineer, taking AI use cases from first workshop to working software in production.
  • Ship agentic systems and automations end to end: agents integrated with client databases, tools, and messaging via the Model Context Protocol, plus voice agents and n8n pipelines.
  • Turn ambiguous business problems into deployed solutions fast, owning scoping, build, evaluation, and iteration with the client in the room.
Progression: Data Science Intern (Feb 2021) → Junior Consultant (Jun 2021) → AI Consultant (May 2024) → Forward Deployed Engineer (Jan 2026)
MAY 2024 – JAN 2026 · LONDON / UK · REMOTE
AI Consultant
Advancing Analytics
  • Design and deploy production LLM, RAG, and agentic AI systems for enterprise clients across insurance, legal, and regulated industries.
  • Integrate AI agents with enterprise systems (databases, design tools, messaging) using the Model Context Protocol, cutting integration from days to hours.
  • Lead AI solution scoping, requirements gathering, and delivery as a Certified ScrumMaster.
  • Evaluate and improve model outputs through systematic prompt engineering, human feedback loops, and quality benchmarking.
JUN 2021 – MAY 2024 · UNITED KINGDOM
Junior Consultant
Advancing Analytics
  • Delivered end-to-end data science projects in Python, SQL, and Azure, from problem framing to deployed models.
  • Built ML models for classification, regression, and NLP, plus dashboards, pipelines, and automated reporting for stakeholders.
  • Advised clients on data strategy, governance, and analytics adoption.
JAN 2019 – MAR 2021 · LAGOS, NIGERIA
Land Surveyor
Gravitas Investment Limited
  • Ran land surveys and geospatial analysis for construction and infrastructure projects using GIS tools.
  • Managed spatial data collection, mapping, and reporting for engineering stakeholders.

Featured Project

PortSwigger AI Pioneer submission
Contract Review Assistant Independent build

AI contract review with a lawyer always in charge.

ProblemIn-house legal teams drown in near-identical contracts. The rare exceptions are what matter, and an AI reading them has to be reliable enough to trust.
ApproachTwo independent model families (Gemini and Claude) extract terms in parallel; a third reconciles them field by field, so agreement becomes evidence and disagreement lowers confidence. Documents are sanitized and personal data is redacted before any model call, and every review is recorded in a tamper-evident audit log.
ImpactCollapses a lawyer's attention from the whole document to the flagged exceptions, each with the offending clause, the playbook rule it breaks, and suggested replacement wording, with a mandatory human sign-off on every outcome.
PythonFastAPILangGraphGemini 2.5 ProClaude SonnetPortkeyReact / TS
Full-stack build: 16-screen React app, streaming batch triage, offline demo mode · live walkthrough on request

Project Showcase · 12 more projects · click any to expand

SurveyEase.ai Independent build AI office suite for land surveyors, live at surveyease.ai.
ProblemLand surveyors do slow, error-prone manual plan production and compliance checking.
ApproachTurns images and coordinates into survey plans, checks them against regulations, and answers regulatory questions with a Claude copilot. Full web and mobile product with auth, payments, and observability.
ImpactA live SaaS that compresses plan production and compliance review, and bridges my surveying background with applied AI.
Next.jsFastAPIPostgreSQLLangGraphClaudeExpoPaystackPostHog
surveyease.ai↗
Under the hoodRedis queues, Cloudflare R2 storage, Sentry error tracking, and Docker/nginx deployment, with an AI gateway managing model routing and retries.
MindEase Pro Independent build An AI wellbeing companion with voice, video avatar, and realtime chat.
ProblemConversational wellbeing AI is high-stakes: it must recognize crisis and respond responsibly.
ApproachVoice and text companion with layered safety checks and region-aware crisis detection, realtime voice, a 3D avatar, and a hardened backend, deployed across Fly.io and Vercel.
ImpactA deployed product where user safety is part of the architecture, not an afterthought.
Next.jsFastAPIGPT-4o RealtimePortkeySupabaseRedisStripeFly.io
Live demo on request
Under the hoodCrisis routing is layered across client and server, and one AI gateway centralizes provider routing, retries, and policy for every model call.
Oracle Independent build A numerology product suite: Telegram bot, web app, and mobile app on one engine.
ProblemOne product across bot, web, and mobile needs a single trustworthy engine, clean entitlements, and reliable payments.
ApproachA versioned, parity-tested reading engine as the single source of truth; a Supabase backend with row-level data isolation and subscription entitlements; web and mobile clients with a custom design system; live Telegram delivery via n8n automation.
ImpactA full consumer product line, from engine to storefronts, built and run end to end by one person.
TypeScriptSupabaseEdge FunctionsNext.jsExpo / RNn8n
Live bot and app walkthrough on request
Under the hoodEntitlements, wire contracts, and the reading engine are shared, versioned packages with parity tests, so every client computes identical readings.
Fifa Arena Independent build A competitive gaming platform with leagues, ladders, and prize contests.
ProblemCompetitive EA FC matches with real entry fees need fair ratings, verified results, and airtight money handling.
ApproachLicensed entry-fee prize contests with a double-entry escrow ledger, Glicko-2 skill ratings, AI screenshot verification of results, and fraud detection, all covered by property-based tests.
ImpactA platform design where fairness and financial integrity are the product.
TypeScriptNext.jsoRPCVitest + fast-checkUSDT settlement
Walkthrough on request
Under the hoodDomain packages encode the escrow and contest state machines and custody rules; property-based tests fuzz the ledger and rating invariants.
EstateIQ Independent build Ask your business database questions in plain English.
ProblemReal-estate teams need answers from their data without writing SQL.
ApproachAn AI analyst that translates plain-English questions into validated, read-only SQL, then returns rows, a chart, and the query itself, plus inventory, collections, and facilities tools in one app.
ImpactNatural-language analytics with every generated query checked before it runs, so answers are safe and auditable.
Next.jsReactTypeScriptSQLiteOpenAI Responses API
Demo on request
Under the hoodA query validator allow-lists tables, permits read-only statements only, enforces row limits, and masks personal data in results.
Zuka Independent build A multi-brand fashion marketplace with an AI shopping assistant.
ProblemIndependent fashion brands need a shared storefront without mixing up each other's data, orders, or payouts.
ApproachA dropship marketplace where brands connect Shopify and a co-branded warehouse reships orders: Stripe and Paystack payments, per-brand data isolation, and a Claude assistant for shopping and product copy.
ImpactA full marketplace across web and mobile with CI/CD and end-to-end tests.
TurborepoNext.jsExpoSupabaseStripePaystackClaude
Deployed on Vercel
Under the hoodRow-level policies scope every record to its brand tenant; Upstash Redis, Sentry, and Playwright/Vitest cover caching, observability, and regression.
Vyra / Afrobeat Studio Independent build An AI music creation studio, live at vyra.app.
ProblemIndependent artists need studio-grade creation and distribution without a studio.
ApproachGenerate, record, and export DistroKid-ready Afrobeats tracks with multi-vendor AI for music and voice, plus built-in AI-disclosure and licensing records so every track is distributable.
ImpactA creator studio spanning mobile, web, and an edge API, with live payments across three providers.
ExpoVite / ReactCloudflare WorkersSupabaseClerkElevenLabsReplicate
vyra.app↗
Under the hoodA pnpm/Turborepo monorepo spanning mobile, web, marketing, and a Hono edge API, with payments across Paystack, Stripe, and Flutterwave.
Hermes Independent build A self-hosted AI operations assistant built as a virtual company.
ProblemRunning several products solo means repetitive operations work that an agent could do, if it can be trusted with keys and money.
ApproachA skills framework organizing 14 departments and 142 reusable skills for a self-hosted agent, with least-privilege keys and human approval required on anything that spends, sends, or signs.
ImpactA blueprint for delegating real operations to agents while keeping people in control.
Agent SkillsPythonMCP
Private
Under the hoodAn org charter defines routing tables and non-negotiable approval floors; a sync script exports all departments and skills as one governed package.
AI Recruiting Automation Independent build An automated hiring pipeline with AI voice interviews.
ProblemScreening and first-round interviews are slow and inconsistent.
ApproachAn n8n pipeline from application form to a Claude resume screen, an ElevenLabs voice interview agent, and Airtable ATS scoring and stage automation, with a white-label Next.js interview app.
ImpactAn end-to-end candidate flow from application to outcome email, human-reviewed at each gate.
n8nClaude (Haiku + Sonnet)ElevenLabs ConvAIAirtableNext.jsWebRTC
62-node live workflow · walkthrough on request
Under the hoodSigned webhooks connect the stages; the interview front end fetches signed URLs server-side so agent credentials never reach the browser.
Regulatory Change Navigator Enterprise A multi-agent watch on regulatory change.
ProblemCompliance teams cannot keep pace with filings and rule changes.
ApproachA LangGraph multi-agent system that monitors SEC filings, detects policy compliance gaps via RAG, and drafts amendments, running on Databricks foundation models with MLflow tracing.
ImpactTurns a manual monitoring chore into an agent-assisted draft-and-review loop.
PythonLangGraphDatabricksMosaic AI Vector SearchFastAPIReact
Hackathon build
Under the hoodLlama 3 70B foundation models with Lakebase/Postgres and MLflow tracing; a FastAPI + React front end surfaces detected gaps for human review.
Speak to PDF Enterprise Talk to your legal documents by voice.
ProblemDense legal and data-protection PDFs are slow to query.
ApproachA voice-enabled RAG assistant over legal and data-protection PDFs, with a Streamlit front end, Azure Durable Functions and Azure OpenAI backend, Azure AI Search retrieval, and a custom speech component I built and packaged.
ImpactDeployed on Azure: ask questions by voice and get grounded, cited answers.
PythonStreamlitAzure Durable FunctionsAzure OpenAIAzure AI SearchAzure Speech
Deployed on Azure
Under the hoodA packaged React/TypeScript speech component handles capture and playback; retrieval is grounded in an Azure AI Search vector index over the document corpus.
Employees Marketplace Independent build A virtual AI company packaged as a plugin marketplace.
ProblemReusable AI capability tends to be scattered and inconsistent.
ApproachA Claude Code plugin marketplace modeling a 14-department company, each department a plugin with its own agents and reusable skills, model-agnostic with an integrity-audit script.
ImpactA composable operating system for agentic work across departments.
Claude Code pluginsAgent SkillsMCPPython
Private
Under the hoodNo hard-coded model IDs; an integrity-audit script validates manifests, and MCP servers for voice, image, and browser are wired in per department.