You will get Build a human-in-the-loop AI agent that proposes, waits for approval
Project details
Most agent demos fall apart in production. An agent that acts autonomously will eventually send the wrong email, delete the wrong calendar event, or push a bad update to your ad account. Autonomous is not a feature, it's a liability.
I build agents with a different architecture. Read live state, propose an action with its reasoning attached, stage the proposal in an approval queue, wait for a human sign-off, execute against the platform API, then verify the change landed and flag any drift. You get an agent that's actually deployable, including to non-technical users and compliance-conscious workflows, plus a full audit trail of every model decision and every human verdict.
The architecture is proven in Aria, my open-source local-first desktop chief-of-staff (daily briefings, email triage, calendar scheduling, meeting capture, RAG Q&A). Every action there goes through approval. See it at github.com/adexdsamson/Aria.
I build agents with a different architecture. Read live state, propose an action with its reasoning attached, stage the proposal in an approval queue, wait for a human sign-off, execute against the platform API, then verify the change landed and flag any drift. You get an agent that's actually deployable, including to non-technical users and compliance-conscious workflows, plus a full audit trail of every model decision and every human verdict.
The architecture is proven in Aria, my open-source local-first desktop chief-of-staff (daily briefings, email triage, calendar scheduling, meeting capture, RAG Q&A). Every action there goes through approval. See it at github.com/adexdsamson/Aria.
AI Development Type
Recommendation System, Software MaintenanceAI Development Language
ScalaWhat's included
| Service Tiers |
Starter
$400
|
Standard
$2,500
|
Advanced
$7,500
|
|---|---|---|---|
| Delivery Time | 5 days | 21 days | 42 days |
Number of Revisions | 1 | 2 | 3 |
AI Model Integration | - | ||
Detailed Code Comments | - | ||
Knowledge Graph | - | ||
Model Documentation | |||
Ontology | |||
Source Code | - | ||
Taxonomy |
Frequently asked questions
1 review
(1)
(0)
(0)
(0)
(0)
This project doesn't have any reviews.
MT
Monochrome T.
Aug 12, 2026
AI Commission & Driver Pay Engine
Good UI screen. Good attitude to feedback given to him and for that reason, He will continue to get better
About Deola
Senior Full-Stack Developer | Next.js, React Native & AI Agents
Lagos, Nigeria - 1:57 pm local time
I recently shipped Aria v1.0, a local-first desktop AI executive assistant built in Electron and TypeScript. Approval-gated tool-calling, hybrid local plus cloud LLM routing, RAG over an encrypted SQLCipher database, and integrations with Gmail, Google Calendar, Outlook, and Todoist.
I also authored MNEMA, an open-source agent memory engine in Python 3.12 with an MCP server, RAG over pgvector, structured LLM outputs, and safety invariants proven with property tests. 124-test hermetic suite, pyright-strict.
Recent production work:
OrbiPayX. Fintech compliance dashboard (React 18, TypeScript, Vite, TanStack Query) with end-to-end AES-GCM encryption and granular RBAC across admins, compliance officers, and support staff. Audit logs with millions of records loading in under a second.
Gidaah. Vacation-rental booking platform on Next.js 15 (App Router) with Stripe payment intents, Veriff identity verification, and SSR/ISR optimization. 50+ cities, 100K monthly actives.
SwiftPro. B2B eProcurement portal (React, TypeScript, React Query, Tailwind, Radix) with role-based navigation for admin, vendor, and evaluator roles, and Playwright end-to-end coverage.
Autogon. Frontend team lead at an AI/ML SaaS: visual ML flow editor with React Flow and React DnD, Stripe subscription management, OAuth (Google, GitHub), Sentry monitoring.
Mobile: cross-platform React Native + Expo apps for fintech (Eapay real-time transactions, 50K+ users; Ulego USD banking, 30K+ users) and a residential-community app (Setraco/Koonage) with 15+ shared components across 20+ modules, biometric auth, and EAS release pipeline.
Core stack:
- Frontend: React, Next.js (App Router, SSR/ISR), TypeScript, JavaScript, Tailwind CSS, ShadCN, Radix UI, React Hook Form, Zod
- Mobile: React Native, Expo, EAS, push notifications, offline caching, biometric auth, App Store and Google Play releases
- Backend: Node.js, PostgreSQL, Supabase, Firebase, REST and GraphQL APIs
- AI: LLM integration (Claude, OpenAI, Qwen), RAG pipelines, vector search (pgvector, sqlite-vec), agentic workflows, tool and function calling, structured LLM outputs, MCP server implementation, prompt engineering
- Payments: Stripe (payment intents, subscriptions), RevenueCat, in-app purchases
- Security: AES-GCM encryption, RBAC, PCI-DSS, OAuth 2.0, JWT, SQLCipher
- Quality: Jest, Playwright, React Testing Library, pytest, pyright-strict, CI/CD, EAS
What I take on:
- Full-stack builds (dashboards, admin panels, SaaS platforms, marketplaces, B2B commerce)
- Next.js production apps with real payments, real auth, and real traffic
- React Native mobile apps with proper release pipelines (EAS builds, App Store, Play Store, RevenueCat)
- AI features embedded in existing products: agents, memory, RAG, structured outputs, approval-gated actions
- Design system implementation, library-first (build the shared components once, assemble every page from them)
- Turning prototypes and AI-generated codebases into production-ready systems
Working style: AI-assisted coding daily (Claude Code, Cursor, Copilot), but I own the output. I care about correctness, clean architecture, and shipping. Comfortable in ambiguity, comfortable owning a whole feature end to end, and comfortable pushing back when a spec is wrong.
Based in Lagos, Nigeria. Fluent English. Available with 4+ hour overlap for US and European time zones.
Steps for completing your project
After purchasing the project, send requirements so Deola can start the project.
Delivery time starts when Deola receives requirements from you.
Deola works on your project following the steps below.
Revisions may occur after the delivery date.
Kickoff and workflow discovery
We map your workflow together on a call. What state the agent reads, what actions it can propose, what a wrong action would break, and where a human sign-off has to land. I leave with a written summary you approve before I write code.
Architecture doc and approval queue schema
You get a written architecture doc scoped to your workflow. It covers the approval queue schema (proposal, reasoning, diff, verdict, verification), the tool-call interface for your platform, and the prompt design for the proposal step.