Javlon Baxtiyorov
← All projects
Prototype  ·  prototype  ·  2025

BioMarketing AI

A marketing agent that turns a product into a quarter's content plan.

A prototype AI agent for Biotact Deutschland GmbH that takes a product brief and returns a full quarter of marketing: a costed content plan, example assets, and targeting — not just a wall of copy.

Role
Author
Impact
Brief → costed Q4 plan in minutes
Status
Prototype
Year
2025
Fig. 8 — Campaign Engine: a product specification feeds an analysis agent that composes a twelve-week content plan and drives it out to channel outlets, metered against targeting and KPIs PRODUCT ANALYSIS AGENT Q4 CONTENT PLAN W1 W6 W12 INSTAGRAM PODCAST EMAIL TARGETING · KPIs fig. 8 — campaign engine PRODUCT SPEC TO COSTED CONTENT PLAN
fig. — biomarketing ai

The problem

Most “AI marketing” tools generate a paragraph on demand; nobody hands the team a whole quarter's plan they can actually schedule and cost.

The useful unit of marketing work isn't a single caption — it's a plan: what to post, on which channel, in which week, aimed at whom, for how much. Most AI tools stop at the caption and leave the planning, costing and targeting to a human, which is exactly the slow part. A caption is a demo; a schedulable, costed plan is a tool. I built for the second one.

What I built

A Streamlit console drives a small pipeline: product analysis, then a structured Q4 plan (~48–60 ideas across Instagram, podcast and email), then example assets with image prompts, then a targeting sheet with demographics, geo, budget and KPIs. The LLM and image APIs live behind placeholder abstractions in `analysis` and `generator`, so a real provider drops in without touching the rest. Output exports to CSV/JSON; a custom logger streams progress into the UI; the plan logic has pytest coverage, and the whole thing ships in Docker. Putting the model behind a placeholder wasn't a shortcut — it's what let me test the pipeline without paying per run.

Key decisions

Make the plan the product, not the prose. Keep the model at arm's length behind an interface so the pipeline can be tested and the provider swapped without a rewrite. It is honestly a prototype — the scaffolding, exports and structure are real; the generation is pluggable rather than wired to a production account. I mark it a prototype on purpose. The archive dates its work honestly.

The result

From one product brief it drafts a 12-week Q4 plan of ~50 ideas across channels, sample assets with image prompts, and a targeting sheet with budget and KPIs — exportable to CSV/JSON.

Brief → costed Q4 plan in minutes

Built with

Python 3.11Streamlit consoleQ4 content planningSwappable LLM / image APICSV / JSON exportPytestDocker

Lessons from the bench

  •  A caption is a demo; a schedulable, costed plan is a tool. I built for the second one.
  •  Putting the model behind a placeholder wasn't a shortcut — it's what let me test the pipeline without paying per run.
  •  I mark it a prototype on purpose. The archive dates its work honestly.

See the code

Open Biotact-AI-Agent on GitHub