LocalLens

AI agent that shows small businesses why competitors outrank them on Google Maps, using live SerpApi data, and gives evidence-backed fixes with English and Telugu WhatsApp reports.

Comment

Pre-existing project: what was added during ForgeHacks (Oct 3-10)

LocalLens is a pre-existing project. The core (SerpApi agent, scoring, fixes, reviews analysis, UI, MCP server) was first built on 27 Sep 2026, before ForgeHacks, for another SerpApi hackathon. The website-audit module (vendor/sitecheck.py) is older still.

Added during ForgeHacks (3-10 Oct, see git history on the repo):

  • Switched the agent from fixture data to live SerpApi mode and ran it end to end on real Google Maps data (Mangalagiri dental clinics).
  • Fixed live-only bugs found in that run: place lookup by place_id, list-typed categories, readable SerpApi API errors, reply-rate fix wording.
  • Real example businesses in the UI, an MCP client that takes a business argument, and a 2:45 demo video recorded on live data.

We are submitting under this disclosure and understand the judges may weigh it accordingly.

Inspiration

India has 60M+ small businesses. For clinics, gyms, coaching centres and shops, a Google Maps "near me" search decides who gets the customer. Owners can see the shop down the road is ahead, but not why, or what to fix first.

What it does

LocalLens is an AI agent that plans, searches, compares and acts on live SerpApi data. Give it "Sunrise Dental Care, Mangalagiri" and it:

  • resolves the business on Google Maps and discovers the competitors Google actually ranks around its coordinates
  • tracks its map rank for three real intent searches
  • reads the latest reviews of the business and its top three competitors (velocity, owner-reply rate, unanswered negatives, praise/complaint themes)
  • audits its website
  • produces a transparent Visibility Score (0-100) next to each competitor's, and a prioritised fix list where every fix cites its evidence (e.g. "you reply to 5% of reviews; competitors reply to 60%")
  • outputs a shareable report and a WhatsApp-ready summary in English and Telugu

A Scan-an-area mode ranks every business in a category and area by fixable gap (for web agencies), and an MCP server (Streamable HTTP) exposes the agent to any AI assistant.

How we built it

Python + FastAPI. SerpApi engines: google_maps (search, place details, location-biased search) and google_maps_reviews. About 9 searches per analysis, with on-disk caching and a hard search budget. Scoring and fixes are rule-based and explainable. Covered by tests; runs without a key on fixtures labelled DEMO DATA.

Challenges

Live-data edge cases that fixtures never showed, and keeping every recommendation tied to real evidence.

What's next

Pilot with local web agencies in Andhra Pradesh, then a paid per-report and agency plan.

AI tool used: Claude (Anthropic) for coding assistance.

Repo: https://github.com/ashokwebs/locallens Demo video: https://youtu.be/IKtz9-5jUOU

Built With

Share this project:

Updates