Devpost Project Story

Inspiration

People do not wake up wanting to configure a marketing stack. They want customers to understand an idea, a local business to become visible, a service to find the right audience, or a new product to reach its market.

That gap is not limited to technical founders. Creators, consultants, nonprofits, operators, small businesses, product teams, and growing companies can all know their work deeply while struggling to coordinate market research, positioning, competitors, channels, content, workflows, budgets, and approvals.

We wanted going to market to feel as simple as configuring a workflow—with the AI doing the substantive work. The customer should be able to start with an idea, add a website if one exists, connect the social identities they own, and review a coherent starting system instead of designing every marketing step themselves.

What it does

Rendara turns a brand or product idea into a cited, reviewable go-to-market workflow.

The customer provides basic context, a goal, a target market, an optional website, and an optional budget ceiling. They connect LinkedIn and X through read-only OAuth. For LinkedIn, they choose a personal profile or company Page before consent; for a company, Rendara then lets them select only a Page they are authorized to manage. For X, they choose personal or brand intent and confirm the exact handle returned by OAuth.

GPT-5.6 then performs the open-ended work: it analyzes first-party evidence when a website exists, researches the market and competitors with native web citations, studies bounded authenticated social samples, proposes an educational content direction, recommends draft workflows, checks a budget allocation, and simulates controlled ad ideas.

Rendara labels every observation as authenticated-provider evidence, public-web evidence, unavailable, or inference. Missing access or metrics remain unknown.

The customer selects one recommended workflow and approves its exact Strategy SHA-256. GPT-5.6 executes that workflow as one LinkedIn asset and one X asset. Deterministic checks verify sources, lengths, workflow binding, review-only state, and the exact Campaign SHA-256 before final approval.

The prototype stops there. It does not publish, reply, monitor communities, activate ads, or spend money.

How we built it

Rendara is a standalone Node.js and Express application built around the OpenAI Responses API, GPT-5.6, hosted web search, and strict structured outputs.

The agentic path is:

idea or brand brief → user-owned social OAuth → optional first-party crawl → cited market/competitor/channel research → authenticated social sampling → GTM synthesis → deterministic strategy checks → exact workflow approval → bounded two-channel creation → deterministic campaign checks → exact campaign approval → stop

The crawler denies private and reserved network destinations, revalidates redirects, respects robots paths, and bounds time, bytes, content type, and page count. A structured model stage must bind every selected competitor's official hostname to a native citation before Rendara samples that domain.

OAuth exchanges happen server-side. State is single-use and session-bound; X uses PKCE. Tokens live only in an ephemeral server-side session store and never enter the browser or generated run. LinkedIn Page selection is restricted to the Pages returned for the authorized member, and X requires confirmation of the exact returned handle.

Canonical JSON and SHA-256 bind both approvals. Application code—not the model—checks request shape, source IDs, evidence classification, provider identity, budget totals and ceiling, disabled live actions, selected-workflow preservation, channel lengths, and exact hashes.

Codex helped us audit the pre-existing Rendara vision, isolate a clean Build Week scope, translate the product promise into explicit contracts, implement and test the agent workflow and OAuth boundary, and challenge unsupported evidence or safety claims. This Build Week prototype does not import the older full platform at runtime.

Challenges we ran into

The hardest product challenge was defining “agentic” without turning it into “uncontrolled.” The AI should do far more than fill a template, but a citation is not proof that every claim is true, OAuth consent is not publishing permission, a budget suggestion is not authority to spend, and campaign approval is not authority to activate.

Those distinctions became visible evidence classes, deterministic checks, identity-bound OAuth state, and two exact approval gates.

The second challenge was social identity. “Connect LinkedIn” is ambiguous when the customer may mean a personal profile or one of several company Pages. Rendara asks first, then uses the authorized identity set returned after consent. X has a different model: a brand is an account, not a separate Page object, so confirming the returned handle is the safest simple interaction.

The third challenge was keeping a broad go-to-market system judgeable in under three minutes. We bounded crawl depth, competitor count, social samples, workflows, ad simulations, source previews, and final assets without reducing the experience to a canned prompt.

Accomplishments that we're proud of

• Anyone with an idea can begin; a website improves evidence but is not mandatory.
• User-owned LinkedIn and X evidence is connected through explicit, identity-aware, read-only OAuth.
• GPT-5.6 performs substantive cited research, synthesis, workflow design, budget planning, and two-channel creation.
• Missing metrics stay unknown instead of becoming confident-looking marketing fiction.
• Strategy and campaign approvals bind the exact reviewed artifacts, not a vague intent.
• The complete prototype produces useful work while exposing no route for a social, ad, CRM, or spend mutation.

What we learned

Agentic products feel most trustworthy when they do more work and make fewer hidden assumptions. The strongest experience is not “AI did everything.” It is: “AI assembled the work, showed its sources and uncertainty, deterministic software checked the contract, and a person approved the exact result.”

We also learned that connection UX is part of agent correctness. If the application cannot tell which human or brand identity authorized the evidence, even a successful API response can be the wrong input.

What's next

The production vision is a one-stop go-to-market workspace with durable brand memory, editable workflows, richer creative formats, activation-specific approvals, scheduled publishing, community-management agents, ad monitoring, experiments, and closed-loop measurement.

Those are future capabilities. This Build Week prototype deliberately proves the safe starting point: idea-to-market research and content creation grounded in user-owned evidence, exact human approvals, and zero external activation.

Short description

Rendara turns a brand or product idea into cited market intelligence, a practical workflow and budget, and approval-gated LinkedIn/X campaign assets. It uses GPT-5.6 for agentic research and creation, user-owned read-only social OAuth for grounded evidence, and deterministic checks for sources, identity, budget, workflow binding, and exact approvals.

One-line pitch

Go from idea to a reviewable go-to-market workflow—with the AI doing the research, strategy, and first campaign draft.

Built With

• openai