← all challenges

// challenge brief · hack a ton 2026

Warm Leads

A small Dobrogea producer declares what they make and how much; the agent finds 3 nearby businesses that should be buying it this week.

// Proposed by
Flavours of Dobrogea
// Industry
Agri-food / B2B sales
// Difficulty
🟡 Intermediate

// the problem

A honey-maker, a small winery, a cheese producer in Dobrogea makes something wonderful, then loses the week trying to sell it one jar at a time at a roadside stand. The real money is in B2B: the restaurant that needs local cheese on its terroir menu, the boutique hotel that wants regional honey in its breakfast baskets, the deli that sells what tourists can't carry home. Producers don't have the time, the contacts, or the sales skill to find those buyers. The buyers are sitting a few kilometers away and never hear about them.

// your mission

Build an agent that turns a producer's profile (declared products, volumes, location, and any free-text notes) into 3 high-quality business leads per week: nearby restaurants, hotels, delis, or shops that plausibly need what they make, each with a reason and a way in.

// how it works

  1. 1Profile: the producer declares products, volumes, price range, delivery radius, and free notes ("organic, 200kg honey/month, can deliver Tuesdays").
  2. 2Discover: build the local universe of candidate businesses from Google Maps (restaurants, hotels, delis, grocers) within the delivery radius.
  3. 3Enrich: go beyond the pin and pull the business type, cuisine, price tier, and where possible the menu so you can infer which ingredients it actually buys.
  4. 4Match: score each producer-business pair on fit (does the menu use honey/cheese/wine?), proximity, volume realism, and demand signals. This is where you innovate.
  5. 5Deliver: hand the producer their top 3 each week with a one-line "why this one" and a draft outreach message, then let them mark hit/miss to sharpen next week.

// what we provide

  • A few real Dobrogea producers' profiles (your demo subjects).
  • A sample target area on the coast to scout against.

// suggested approach

  • Inference: for model access, we suggest [LLMok](https://llmok.app). Use code AMBASADA26 for 50% off.
  • Pull candidate businesses for an area from Google Maps, then enrich each one. Menu text (from a website, an aggregator listing, or a photo) is the signal that turns a generic "restaurant nearby" into "this kitchen buys goat cheese."
  • The match logic is the scored bar: an LLM reasoning over product + menu + business type beats a keyword filter. Show your reasoning per lead.
  • Floor version: one producer + one area → a ranked list of candidate businesses with a fit reason. Add menu enrichment, weekly cadence, and feedback learning from there.
  • On data: scraping breaks platform terms and rate limits will bite. Prefer the official Places API, partner-provided lists, or a small hand-built seed set for the demo, and be honest in the pitch about how you'd scale collection.

// how you'll be judged

  • Quality of the matches: would a producer actually call these 3 businesses, or is it a random nearby list? (the scored bar)
  • The matching intelligence: how well menu/business signals map to a producer's products.
  • Usefulness of the "why" and the draft outreach.
  • The agentic loop: weekly cadence and learning from hit/miss feedback.
  • The database angle: does the system build a reusable, self-updating map of local producers and buyers?

// stretch goals

  • Menu-level demand inference ("this bistro lists 4 cheese dishes and no local supplier credited").
  • Two-sided value: a business view that surfaces producers who can supply it.
  • Freshness: detect when a business closes, changes its menu, or a new one opens.
  • Outreach that closes the loop: draft the email/WhatsApp and track replies.

// deliverables

  • Git repo: code + README + run instructions.
  • Demo: a real producer's 3 leads for the week, each with a reason, on screen.
  • 5-minute pitch: architecture, the matching approach, decisions, trade-offs.