LeadRadar
An open-source AI agent tool that automatically searches local businesses on Google Maps, scores leads, and drafts personalized outreach emails, helping freelancers find clients efficiently.
Timeline (last 90 days)
LeadRadar is an open-source AI agent project submitted by developer mudit-palvadi in the second week of the Hacktoberfest open-source AI challenge, deployed on Render, under the MIT license, with 68 tests.
LeadRadar's function is: based on the occupation, city, name, and desired number of leads entered by the user, it automatically searches for local businesses via Google Maps, scores and ranks the results, and drafts a signed outreach email for the top-ranked lead.
LeadRadar's tech stack is: Express for the API, MongoDB Atlas for storage, the allam-2-7b model on Groq's free tier (with Ollama as a local fallback), SerpApi for search, all deployed on a single Render service.
LeadRadar's scorer downweights the model score to 25%, with the remaining 75% coming from a deterministic weighted formula based on search facts, for example 55 points for having no website, 15 points for a low rating with no replies to reviews, and a base score of 30.
LeadRadar requires the search to be generated automatically by the agent, with the user not typing query terms by hand: the agent first asks the model which local businesses would hire that occupation, then assembles a Google Maps query.
When designing LeadRadar, the creator hit a pitfall: the initial prompt contained only one web design example, causing the agent to recommend irrelevant businesses such as toy stores, pet stores, car washes, and bookstores to a graphic designer; after rewriting the prompt to add cross-occupation examples, deduplication, and similar-occupation filtering, the same Austin scan produced 15 businesses worth outreach.
LeadRadar uses isUsableResult() to filter out non-business records returned by SerpApi (most commonly results that return only a city name with no real title), preventing the model from fabricating details based on them.
LeadRadar's email layer performs output validation in code: ensureSignOff() fills in sign-offs the model omitted, cleanEmailBody() removes sentences where the model says it did not check the website and prohibited clichés, and pitchGuidance() prevents the pitch from mismatching the occupation, rather than relying only on the prompt.
The author's design rule of thumb is: if something can be solved by checking the output in code, don't fight with the model in the prompt.
LeadRadar runs on Render's free tier, and the first visit after a cold start takes about one minute.
1 more timeline events
All Facts (11)
LeadRadar runs on Render's free tier, and the first visit after a cold start takes about one minute.
50%UnverifiedLeadRadar is an open-source AI agent project submitted by developer mudit-palvadi in the second week of the Hacktoberfest open-source AI challenge, deployed on Render, under the MIT license, with 68 tests.
50%UnverifiedLeadRadar's tech stack is: Express for the API, MongoDB Atlas for storage, the allam-2-7b model on Groq's free tier (with Ollama as a local fallback), SerpApi for search, all deployed on a single Render service.
50%UnverifiedLeadRadar's scorer downweights the model score to 25%, with the remaining 75% coming from a deterministic weighted formula based on search facts, for example 55 points for having no website, 15 points for a low rating with no replies to reviews, and a base score of 30.
50%UnverifiedLeadRadar requires the search to be generated automatically by the agent, with the user not typing query terms by hand: the agent first asks the model which local businesses would hire that occupation, then assembles a Google Maps query.
50%UnverifiedLeadRadar's function is: based on the occupation, city, name, and desired number of leads entered by the user, it automatically searches for local businesses via Google Maps, scores and ranks the results, and drafts a signed outreach email for the top-ranked lead.
50%UnverifiedLeadRadar uses isUsableResult() to filter out non-business records returned by SerpApi (most commonly results that return only a city name with no real title), preventing the model from fabricating details based on them.
50%UnverifiedLeadRadar's email layer performs output validation in code: ensureSignOff() fills in sign-offs the model omitted, cleanEmailBody() removes sentences where the model says it did not check the website and prohibited clichés, and pitchGuidance() prevents the pitch from mismatching the occupation, rather than relying only on the prompt.
50%UnverifiedThe author's design rule of thumb is: if something can be solved by checking the output in code, don't fight with the model in the prompt.
50%UnverifiedBefore release, two freelancer friends tested LeadRadar: the web developer said he liked the app and would keep using it, but did not convert; the graphic designer said the app could not find good leads suited to his occupation and felt it was more optimized for other skills.
50%UnverifiedWhen designing LeadRadar, the creator hit a pitfall: the initial prompt contained only one web design example, causing the agent to recommend irrelevant businesses such as toy stores, pet stores, car washes, and bookstores to a graphic designer; after rewriting the prompt to add cross-occupation examples, deduplication, and similar-occupation filtering, the same Austin scan produced 15 businesses worth outreach.
50%