Direct contractMulti-agent AI SaaS · 2025–2026 · Founding engineerapp.lainey.io
Lainey AI
Founding engineer on a multi-agent AI platform. Architecture, calendar integrations and automated content scheduling, shipped solo.
Lainey AI is a direct Upwork contract: 500 hours as founding engineer on a multi-agent AI SaaS product, covering architecture, Google Calendar and Apple Calendar integration, and automated content generation and scheduling, rated 5.0 across a contract still running into 2026.
- Direct
- Attribution
- 500 hrs
- Scope
- 5.0
- Status
500 hours, $8,713.33, 5.0 rating, public on Upwork, Jun 2025–Apr 2026.
The problem
Lainey AI had a product thesis and no engineering team. The product needed to manage a user’s tasks across Google and Apple calendars and generate and schedule content on their behalf, which meant real agent architecture, real authentication against two different calendar providers and real scheduled execution rather than a prompt in a text box.
The founding engineering role covered all of it: system design, integrations, authentication, deployment, and the judgement about what to build first.
Constraints
The conditions the solution had to survive, not a wish list.
- One engineer
- Architecture through production owned by a single developer. Anything too complex to operate alone was the wrong answer.
- Third-party quotas
- Google Calendar and Apple Calendar run on different auth, sync and rate-limit models, which shape the design rather than sit at the edge of it.
- Non-deterministic core
- Model output varies run to run. The system around it had to stay predictable when the model did not.
- Token cost
- Every agent step carries a direct dollar cost, making context strategy a budget decision as much as a quality one.
The approach
In plain English, before the architecture.
The build treats the language model as one component inside a conventional system, not as the system itself.
Work is split across specialised agents with narrow responsibilities and explicit handoffs, which keeps each prompt small, cheap and testable. A planning step decides what needs to happen; execution steps each carry out one job against a real API.
Everything that touches the outside world runs through ordinary application code with validation, retries and error handling, so a bad model response becomes a handled failure rather than a corrupted calendar.
Architecture
The decisions that mattered, and why they were made that way.
- 01
Multi-agent orchestration
Specialised agents handle planning, retrieval and execution with structured handoffs. Narrow scopes keep context windows small, cut token spend and make each step independently verifiable.
- 02
Schema-constrained model output
Model calls return structured output validated before it reaches any side effect. Anything failing validation is retried or surfaced as an error rather than acted on.
- 03
Google Calendar and Apple Calendar integration
OAuth with scoped, refreshable access against Google Calendar, and CalDAV-based sync against Apple Calendar, each with its own auth and refresh model, plus graceful degradation when a user revokes access or a sync window fails.
- 04
Automated content generation and scheduling
An execution agent drafts content on the user’s behalf and places it directly onto their calendar as scheduled events, so the output lands where the user already works rather than in a separate feed to check.
- 05
Scheduled execution on AWS Lambda
Recurring agent work runs on scheduled Lambda invocations rather than a long-lived process, keeping idle cost near zero and making every run independently retryable.
Outcome
Delivered as the sole founding engineer, from an empty repository through architecture, integrations and production deployment. In the client’s own words, “an excellent communicator and outstanding at what he does.”
The contract is still active into 2026 on the strength of that build, rated 5.0 across its full duration.
Stack
- Next.js
- React
- Node.js
- OpenAI
- Google Calendar API
- Apple Calendar (CalDAV)
- AWS Lambda
Timeline
June 2025 to April 2026, 500 contracted hours
Contract value
$8,713.33
More work
BlackAlgo
Algorithmic trading platform
Full build
Avana
Multi-tenant booking SaaS
Full build
MyFileTracker
EU B2B e-invoicing
13 days
Tell us what you can’t staff.
Twenty minutes. No deck, no discovery sequence. If we are the wrong fit we will say so on the call and point you somewhere better.