Fundam

Fundam uses AI to assess what funding a small/medium enterprise (SME) can access, identify gaps, match suitable opportunities, and create the documents needed to submit strong funding applications.

Comment

Inspiration

Fundam was inspired by a critical economic gap: viable small businesses struggle to access capital even when funding opportunities are actively available.

In emerging markets like Nigeria, the primary bottleneck is rarely a total absence of capital. Grants, loans, and equity funds exist, but small and medium enterprises (SMEs) struggle to determine their readiness, identify matching criteria, or translate informal operational records into standard financial documentation required by institutional lenders.

These businesses regularly generate transaction data through bank statements, sales receipts, and day-to-day records. Fundam was built to turn these raw operational footprints into actionable financial readiness.

What it does

Fundam operates as an AI-native intelligence layer between SMEs and capital providers. The platform automates three core functions:

  • Readiness Assessment: Evaluates raw financial records to calculate real-time funding readiness scores.
  • Gap Identification: Pinpoints specific compliance, record-keeping, or liquidity deficits holding the business back.
  • Automated Application Preparation: Matches qualified businesses to suitable capital routes and automatically generates pitch decks, cash flow projections, and application packs.

How we built it

Built from scratch starting in July 2026, Fundam was designed around Gemini agents and Google Cloud from day one — not bolted on later. We used the Gemini ecosystem end to end: Antigravity to scaffold the first MVP and agentic coding workflow, Stitch to prototype founder-facing screens (questionnaire, dashboard, document upload, packs), then hardened that into a production system.

Stack. The backend is Python / FastAPI — agents, orchestration, payments, OCR, pack export, and matching. The frontend is TypeScript / Next.js / React with Tailwind, talking to the API for onboarding, readiness results, and pack download. Auth is Firebase; payments are Paystack (₦ pricing for Nigerian SMEs). Services ship in Docker to Google Cloud.

Google Cloud infrastructure. We run three Cloud Run services: public API, frontend, and a private worker that executes the AI pipeline. Cloud Tasks queues async jobs so scoring and pack generation don’t block the UI. Uploads and generated packs live in Cloud Storage. App state and opportunity embeddings sit in Cloud SQL (PostgreSQL + pgvector). Secrets go through Secret Manager.

AI layer. Document AI OCRs messy SME PDFs/images. Gemini on Vertex AI (default gemini-2.5-flash) structures metrics, scores readiness, recommends funding routes, runs advisory funder-fit judgment, and drafts editable packs — pitch deck (PPTX), financials, checklists. Google ADK shapes the multi-agent orchestration path. Opportunity matching uses Gemini embeddings queried in-database via pgvector.

Cost lesson that shaped the architecture. A full staging environment with Gemini services drove cost up quickly. We migrated matching off dedicated vector-search serving capacity onto pgvector on Cloud SQL, keeping Gemini for embeddings and reasoning while cutting idle infra burn — so AI-native ops stay sustainable as assessment volume grows.

Production pipeline (what runs when a customer pays). Questionnaire → payment unlock → Document AI OCR → Gemini extract & financial narrative → compliance / advisory Gatekeeper → funder-fit judge → pack generation → opportunity matching.

All of that is executed by Cloud Tasks → Cloud Run workers hitting Vertex/Gemini in production. Founders edit and download the outputs; admins approve discovered funders before they enter the matching corpus.

Result: a real full-stack product — Python agents on Google Cloud, Next.js product UI, Gemini live on every paid assessment — built to turn messy SME inputs into funder-ready materials at SME-accessible price points.

Challenges we ran into

Unstructured SME data: Bank statements and ops records arrive as messy PDFs/images, not clean ledgers. We iterated hard on Document AI + Gemini extraction and parsing so we could standardise varied formats without dropping critical ₦ figures or context.

Generating useful packs, not empty shells: Early drafts were thin or generic — the docs founders need (pitch deck, financials, checklists) weren’t “there” until we invested in stronger prompts, templates, and length/grounding rules so outputs were realistic, editable, and grounded in their answers (with clear “to be confirmed” where data was missing).

Calibrating trust vs speed: AI can score and draft fast; capital decisions can’t be faked. We kept Gatekeeper/funder-fit advisory, refused invented metrics, and avoided false “you’re approved” language so readiness reports stay honest about gaps and true capital requirements.

Security & trust for financial uploads: Sensitive docs needed explicit consent, secure storage on GCS, and a clear boundary: Fundam organises and drafts — it does not claim to hold or verify CAC/tax/ID as if it were the funder.

Accomplishments that we're proud of

Commercial validation in weeks: Converted real Nigerian SME users into paying customers within the first month of launch, proof the product solves a willingness-to-pay problem, not just a demo use case.

AI that produces the product: Turned weeks of manual advisory work (financial read of messy docs + funding application packs) into a live agent pipeline, Gemini scores readiness, structures evidence, and drafts pitch decks, financials, and checklists customers actually download.

Built for real SME realities: Designed for raw operational uploads and incomplete books not the assumption that every founder already has formal accounting, polished decks, or a consultant on retainer.

Production AI without runaway infra cost: After staging Gemini bills spiked, we migrated opportunity matching to pgvector on Cloud SQL — keeping Gemini for embeddings and reasoning while cutting the need for dedicated vector-search serving capacity.

Google Cloud in production: Shipped the full path on Cloud Run, Cloud Tasks, Vertex/Gemini, Document AI, GCS, and Cloud SQL, AI running on every paid assessment, not only in the pitch video.

What we learned

  • Pathfinding Over Discovery: SME owners do not just need directories of funding sources; they need direct interpretation of whether an opportunity is right for them and step-by-step guidance on what to do next.
  • Proven Willingness to Pay: Converting early users into paying customers proved that funding readiness is a high-intent, immediate pain point for growing businesses.

What's next for Fundam

We are building Fundam to become the primary financial infrastructure bridging informal business operations with formal capital markets across Africa, expanding from Nigerian Creative Industry SMEs to all Nigerian SMES across sectors and then into regional markets.

Built With

Share this project:

Updates