MandiLens
A multilingual decision-support product that turns official AGMARKNET reports into market histories, comparisons, arrivals analysis, and short-horizon outlooks.
The product, the pressure, and the response.
Overview
MandiLens is an analysis layer over official AGMARKNET reporting. It helps people inspect market evidence, but it does not replace the official source or local market knowledge.
Context
Daily mandi reports contain useful price and arrivals information, but inconsistent naming, reporting gaps, and variety-level rows make direct comparison difficult. A useful product first needs a traceable cleaning and validation workflow.
Solution
A batch-first Polars pipeline inspects and prepares market-day observations, evaluates short-horizon methods with chronological data, and exports a compact product artifact. The Next.js interface makes the result searchable and available in 10 languages without a runtime model API.
Founder and lead developer
Source inspection, data engineering, model evaluation, product engineering, multilingual delivery, and deployment.
Market histories
Search prepared price and arrivals observations by state, commodity, market, and reporting period.
Comparable market views
Review markets and seasonal patterns from normalized market-day observations.
Decision inputs
Use quantity and entered transport-cost assumptions to compare estimated net realization locally in the browser.
Evaluated outlooks
Present short-horizon model outputs with methodology and data-quality context rather than treating them as guaranteed prices.
Multilingual access
Navigate the public product in 10 supported languages.
Built in reviewable steps.
- 01
Inspect source reports
Profile AGMARKNET records and preserve source lineage before cleaning or aggregation.
- 02
Prepare observations
Use Polars to normalize names, validate ranges, handle reporting gaps, and create market-day records.
- 03
Evaluate chronologically
Assess scikit-learn methods against time-ordered data so future observations do not leak into earlier evaluation periods.
- 04
Export the public product
Publish a compact data artifact for a responsive Next.js interface with no runtime inference service.