ML Engineer @ Adludio — LLM backends & Lambda APIs (campaign automation)
Lorenzo — AI Campaign Strategy Builder
At Adludio: a chat-first campaign builder so advertisers go from rough idea to structured, pre-targeted drafts without long static forms on the mobile ad stack.
Project gallery
Use the side arrows to browse; click an image for full size.
Faster setup for advertisers; campaign managers spend less time cleaning briefs once campaigns are in flight.
Lorenzo — AI Campaign Strategy Builder
Adludio (London), ML Engineer. Lorenzo streamlines campaign creation for advertisers: instead of wrestling with rigid forms, they describe goals in chat and the product assembles a coherent campaign strategy—brief, segments, and suggested targeting—before anything goes live.
Scope: Backend & AI integration — LLM prompt design, structured outputs, AWS Lambda APIs, and wiring into the existing campaign UI.
The problem
Campaign setup relied on a long static form. Advertisers often submitted thin or inconsistent briefs; problems showed up only after campaigns were running. Campaign managers then chased clients for clarification—slow for clients and expensive for the team.
What I built
A chat-first campaign builder that automates the early setup path:
- Guided brief capture. Natural-language messages are checked in real time; the model rejects vague or off-topic input and asks for specifics so the campaign definition stays usable.
- Structured strategy. From a short description it infers industry, product context, and implied audience (e.g. American football in Europe vs generic “sports”).
- Targeting suggestions. Age bands, interest groups, and keyword clusters are proposed as editable defaults, then flow into the platform’s targeting model when accepted—so creating the ad campaign is faster and more consistent for clients.
Architecture
- LLM layer — Prompt templates with strict schemas for validation, context extraction, and suggestions; retries when the model violates the schema.
- API — Serverless Python on AWS Lambda for the chat flow, session state, and stable JSON contracts for the frontend.
- Integration — Embedded in the existing management UI so approved segments land directly in the live campaign configuration.
Outcome
Less back-and-forth after onboarding: advertisers moved from idea to a validated, pre-targeted draft campaign in one session instead of email threads and form revisions. Campaign managers spent less time cleaning bad briefs post-launch.