Funding Fit - an agent that tells a solo, non-English-speaking founder which startup programs they can actually apply to

Description and use

Funding Fit is an agent that helps solo founders without a company and without English video interviews determine which startup programs they can apply to, using a local LLM and a Sanity knowledge base.

Problem

Solo founders without a company, staying in Korea, and unable to do English video interviews struggle to determine which startup programs they can apply to because eligibility information is scattered across program pages.

Outcome

The agent identified Hustle Fund, TheVentures, FuturePlay, and Mashup Ventures as programs with no blocking eligibility rules for the founder’s profile, Antler Korea as conditional, and excluded Y Combinator, Founder University, SparkLabs, and Betaworks Camp.

Workflow

• 1. Create a Sanity knowledge base with 14 official program sources organized into 9 entry categories.
• 2. Define a purpose field to focus on eligibility, residency, interview format, and application process.
• 3. Deploy a local model (gemma3:4b via Ollama) to run all LLM calls locally.
• 4. Use the Context MCP endpoint (funding-fit) with read‑only Context Viewer token.
• 5. Agent workflow: initial_context → knowledge_base_search → knowledge_base_read → extract 8 fields per program with quotes.
• 6. Grounding check: verify each value’s quote appears in the retrieved text and matches the topic.
• 7. Compute fit verdict (apply / conditional / excluded) and final summary from verified facts only.
• 8. Present results in Korean and English.
• 9. Record performance metrics (e.g., 47 facts extracted, 39 verified, 100 s runtime).

Evidence

Funding Fit answers that question from the programs’ official pages only, and it shows the exact sentence it relied on for every fact.

Source

All model calls run locally with a small open model (gemma3:4b via Ollama) - free, offline, nothing about me leaves the PC except the search query to Sanity.

Source

The fit verdict (apply / conditional / excluded) and the final summary are computed by code from verified facts only.

Source