Founding AI Product Engineer

Parker AI

Building the AI-native creative-strategy platform for high-spend Meta advertisers.

Timeline

06/2023 – Present

Location

Remote · Berlin, Germany

Stack

4 areas · 20 tools

Scope

5 areas · 7 highlights

Overview

Parker AI is an AI-native creative strategy and marketing intelligence platform that helps consumer brands and DTC teams turn raw social and ad signal into shippable creative.

As founding AI Product Engineer I own the platform end-to-end: data ingestion at scale, retrieval and embedding pipelines, agent orchestration, AI tool-calls, the Slack-first product surface, and the analytics that brands act on every day.
Customers spend up to $2M per month on Meta ads and rely on Parker to find what to make next — winning hooks, scripts, angles, and full creative briefs.

My role

Architect AI pipelines from data ingestion through retrieval, generation, and human review.
Design and ship multi-agent workflows that plan, call tools, write to memory, and produce auditable outputs.
Build the Slack-first surface and Next.js dashboard that ops, strategists, and brand teams live in.
Run product-grade evals and observability on every model call so quality is measurable, not vibes.
Partner on AI unit economics: prompt budgets, retrieval/caching strategy, and quality gates that reduce wasted inference.

Stack

AI & Orchestration

MastraTemporalMCP-style toolsLangfuseOpenAI / Anthropic / Gemini

Backend & Data

Next.jsNode.js / TypeScriptSupabase PostgresQdrantRedis

Infra

GCPCloud RunCloud FunctionsCloud Storage

Integrations

Meta Marketing APITikTokInstagramRedditSlackPostmark

Highlights

Built a multi-source ingestion pipeline pulling TikTok, Instagram, Facebook, Reddit, competitor sites, reviews, and Meta ad performance with robust schedulers, retries, and dead-letter queues on Temporal.
Designed a hybrid retrieval layer combining Qdrant vector search with Supabase Postgres relational data, including chunking, embedding serialisation, and query-aware re-ranking.
Shipped agentic ideation systems on Mastra that generate hooks, scripts, angles, briefs, and a reusable idea bank — with tool calls into ads, analytics, competitor research, and creative review.
Built Slack-first delivery: proactive alerts, weekly strategist reports, ad-performance digests, and creative recommendations that surface where teams already work.
Implemented Meta Marketing API integration with advanced permissions (ads_management, pages_read_user_content), token rotation, and rate-limit-aware fetchers.
Instrumented every LLM call with Langfuse traces, prompt versions, evals, and cost/latency dashboards so the team ships changes with confidence.
Drove cost/perf trade-offs with model routing and cache-aware retrieval so teams could scale usage without sacrificing response quality.

Outcomes

Powering creative strategy for brands at up to $2M/month Meta ad spend.
Cut creative-idea turnaround from days to minutes with an always-on idea bank.
Built the foundation other engineers now extend — pipelines, agent rails, and eval harnesses.

FAQ

What does the Parker AI stack look like end-to-end?

Next.js for the dashboard, Node.js services on GCP Cloud Run, Temporal for durable workflows, Supabase Postgres + Qdrant for hybrid retrieval, Redis for hot state, Mastra for agent orchestration with MCP-style tool calls, and Langfuse for tracing, evals, and prompt management.

How is data ingested from TikTok, Instagram, and Meta?

Temporal workflows run schedulers and fan-out workers that pull from each platform, normalise into a canonical schema, embed long-form content for vector search, and write both relational and vector representations. Failures are retried with backoff and dead-lettered for review.

How are AI agents kept safe and predictable in production?

Every tool call is typed, traced through Langfuse, and bounded by deterministic rails. Agents have explicit memories, budget caps, and human-in-the-loop checkpoints on high-impact actions like sending strategist reports.