Asar Marketing Intelligence

An AI that is specialised in suggesting marketing campaigns. The model forces the AI agent to think critically, empathise with users, ask questions, and suggest marketing campaigns using a workflow.

Comment

Inspiration

During the first two years of my entrepreneurial journey, I identified around 15 to 20 problems that I believed were worth solving. I could often imagine a solution, but I did not have the money, team, or development resources to turn those ideas into products.

At the same time, I noticed another problem. Founders were asking general AI tools for marketing help, but the answers often felt generic:

  • Post consistently
  • Work with influencers
  • Run a giveaway
  • Create engaging content

These answers may sound reasonable, but they are often given without understanding the product, audience, budget, previous failures, conversion path, or why a customer would care.

That is not strategy. It is guessing.

When OpenAI launched GPT-5.6 Sol, I wanted to test whether it could give me enough development leverage to finally build one of my ideas.

That idea became Asar Marketing Intelligence, an AI marketing system that understands the business before recommending how to market it.

GPT-5.6 Sol did not create the idea behind Asar. It gave me the development leverage to turn that idea into a working product.

What it does

Asar helps founders develop, challenge, test, and execute marketing campaigns based on the reality of their business.

If someone says:

“I run a small café. Help me get more repeat customers.”

Asar does not immediately create a generic campaign. It first asks focused questions about the café, customers, budget, current marketing, previous attempts, and desired result.

Once it has enough information, Asar creates a practical campaign containing:

  • The campaign idea and hook
  • Why a real customer would care
  • Execution steps
  • Call to action
  • Conversion path
  • Risks and guardrails
  • KPIs and success criteria

Asar can also:

  • Challenge weak marketing ideas
  • Run reality checks on its own recommendations
  • Compare different campaign directions
  • Research current campaigns and market signals
  • Learn principles from successful campaigns without copying them
  • Test campaign assumptions
  • Remember approved business information
  • Turn campaigns into dated marketing tasks
  • Keep the founder as the final approval authority

The main innovation is the workflow around the AI. Asar asks before it assumes, checks before it recommends, and connects every campaign to a measurable business result.

How we built it

We built Asar through 13 Codex conversations, mainly using GPT-5.6 Sol at medium reasoning effort.

The application uses:

  • React and Vite for the frontend
  • Node.js for the backend runtime
  • DeepSeek V4 Pro for marketing reasoning
  • DeepSeek native web search for current research
  • Supabase for authentication, PostgreSQL storage, and user-level data protection
  • Markdown and YAML for marketing protocols and auditable memory
  • Server-Sent Events for streaming AI responses
  • Vercel for deployment

Asar has a reference library containing campaign case studies and reusable marketing patterns. It retrieves only the resources relevant to the current request.

The memory system separates confirmed business facts from temporary discussion. This helps Asar remember useful information without treating every brainstorming idea as an approved decision.

Challenges we ran into

One of the biggest challenges was stopping the AI from making silent assumptions.

Asar needed to ask enough questions to understand the business, but it could not turn every conversation into a long questionnaire. We built an adaptive interview flow where each question depends on the information still missing.

Memory was another challenge. We needed to keep confirmed facts separate from suggestions, assumptions, and working discussions. If a user changes an approved detail, Asar must replace the old information instead of remembering two conflicting versions.

Live research was also difficult because marketing signals change quickly. Asar needed current sources while still treating online information carefully and avoiding the direct copying of existing campaigns.

Performance created another challenge. Deep campaign work can require research, planning, review, and a final answer. We had to control context size, reduce unnecessary AI passes, stream progress, and show clear failures instead of leaving the user waiting.

Finally, we had to turn marketing advice into action. A campaign idea is incomplete if the founder still has to manually create the tasks, timeline, responsibilities, measurements, and approval process.

Accomplishments that we're proud of

We are proud that Asar became a complete working product instead of remaining an idea.

Some of our main accomplishments include:

  • Building an adaptive AI that asks questions instead of following a fixed questionnaire
  • Creating campaign recommendations based on real business context
  • Adding live marketing research with visible sources
  • Building a memory system that separates confirmed facts from temporary discussion
  • Giving Asar the ability to challenge both user ideas and its own recommendations
  • Turning saved campaigns into dated execution plans
  • Keeping external publishing, spending, and customer contact behind founder approval
  • Deploying a secure multi-user version with protected account data
  • Building and testing the project across 13 Codex conversations

We also ran a small internal comparison between direct model answers and answers produced through the Asar workflow. The results suggested that a strong workflow could improve the usefulness of the same underlying model. This was a limited internal experiment, not a scientific benchmark.

What we learned

The biggest lesson was that the AI model is only one part of an AI product.

A powerful model can still give a weak answer if it receives incomplete context. Better questions, controlled memory, focused research, honest reality checks, and clear execution can make the final answer much more useful.

We also learned that users should be able to see what the AI is doing. Research sources, memory updates, actions, progress, and failures should be visible instead of hidden.

Most importantly, we learned that AI should not replace the founder’s judgment. It should help the founder understand the situation, question weak assumptions, and make better decisions.

Asar is not here to give founders more ideas. It is here to stop AI from confidently solving the wrong problem.

What's next for Asar Marketing Intelligence

The next step is to launch Asar as a complete SaaS product with a paid plan.

We plan to add more features that connect campaign strategy, research, testing, execution planning, performance measurement, and business memory in one product.

The final version of Asar will be powered by both DeepSeek V4 Pro and GPT-5.6 Sol. Each model will be used where it provides the most value across research, reasoning, campaign development, review, and execution planning.

Our goal is to make Asar a complete AI marketing intelligence platform for founders and growing businesses.

Built With

Share this project:

Updates