FounderLens AI — Evidence before decisions
Current status: Concept and submission draft. A working agent, final model/provider selection, public code, demo video, and measured results are not yet available. The workflow below describes the intended product, not completed functionality.
Inspiration
Early-stage founders spend valuable time gathering market signals, comparing alternatives, and turning scattered information into a decision-ready brief. An AI agent should reduce this preparation work without pretending to know whether a startup idea will succeed.
What we're building
FounderLens AI is a proposed evidence-first research assistant for one concrete job: helping a founder investigate a business assumption before choosing the next customer-discovery step. The goal is a short, checkable brief that distinguishes observations, inferences, unanswered questions, and suggested experiments.
Planned user journey
- The founder enters an assumption, target audience, market, and research question.
- The agent collects relevant public or user-provided information and records source links and dates where available.
- A free-access or open-weight language model drafts a concise evidence map: supporting signals, contradictory signals, limitations, and interview questions.
- The founder reaches an explicit human checkpoint: inspect citations, challenge claims, edit hypotheses, and approve or reject recommendations.
- Only the human-approved brief can be exported. No emails, lead outreach, purchases, or other external decisions are made automatically.
Human Judgment by Design
The agent prepares: source summaries, assumption maps, uncertainties, and first-draft research questions.
The founder decides: whether sources are credible, what evidence is missing, which hypothesis to test, what to change, and whether to export a brief. An attractive recommendation is never treated as an instruction to act. Rejected claims remain identifiable for review.
Proposed architecture
Founder input → research orchestration → public/user-provided source collection → free-access/open-weight model → claim/source validation → evidence review interface → human approval → export.
This is a proposed architecture, not an implemented or tested system diagram. The chosen model, hosting, libraries, data sources, deployment, and fallback behavior will be documented after implementation.
Equal Access AI and disclosure
The planned runtime is restricted to models available to all participants at no cost, through a genuinely free tier or an open-weight local option. No paid runtime credits or subscriptions are planned. The final AI Disclosure Statement will identify each actual model, provider, tool, access plan, role, and an observed or estimated call/token count per run after testing. AI assistance used to design or implement the project will be disclosed.
Time-Saved Report — measurement plan
For representative founder research tasks, record the same output requirements under two conditions: manual research and FounderLens-assisted research, including human review and corrections. Track active minutes, source accuracy, correction effort, and completed briefs. Report raw observations and the median time difference; do not count a time saving until data is collected. No time-saving percentage is claimed at this stage.
Responsible AI
The intended demo will use public or synthetic business scenarios. We will avoid uploading confidential founder data, personal information about prospects, or credentials. Known risks include hallucinated citations, stale market signals, geographic/language bias, omission of negative evidence, and overconfidence. Mitigations to test include source links, visible uncertainty, explicit unsupported-claim labels, prompt-injection-aware retrieval, data minimization, and a mandatory human checkpoint.
Challenges and next steps
The central design challenges are finding verifiable evidence within free runtime limits, making uncertainty understandable, and proving that founder review actually improves decisions. Upcoming work includes implementing the prototype, selecting and documenting the free model, testing the full workflow, measuring time saved, preparing the architecture diagram, and recording a demo of no more than three minutes. The team is still being formed.
Submission note: The repository, live test link, measured report, and demo video will be added only when real artifacts exist and have been verified.
Concept walkthrough video (not a functional demo)
The linked YouTube video is a 2-minute-6-second illustrative product concept walkthrough showing planned screens and the proposed human-approval checkpoint. It does not depict a running AI agent, tested automation, verified research results, or measured time savings. A separate real end-to-end demonstration will replace it once implementation and testing are complete.
