JOY Automart's combat lakehouse. ARES crawls Bangladesh's auto-parts market continuously — Daraz, Bikroy, Pickaboo, local supplier WhatsApp groups — builds historical pricing graphs, surfaces predatory drops, and recommends margin-preserving counters before the rival pages even cache. Named for the war god — every move is calculated to win.
Protect the margin moat. Surface every competitor price war 48h before it lands. Recommend the surgical counter — undercut, hold, or pull from the war altogether — and feed JOYADH a decision-ready brief.
Daraz · Bikroy · Pickaboo · 27 known local supplier groups. Pricing graphs refresh every 2h.
Historical cadence model flags suspicious rival pricing 12-48h before they discount.
What if we drop 8%? ARES simulates net margin given landed cost (from ONI) + FX (from AAJ).
One-page JOYADH-ready brief: rival SKU · price delta · suggested response · margin floor.
Every JOY SKU mapped to its top 3 rivals across all watched marketplaces.
Detects when a rival is selling below cost (loss-leader to grab market share) and recommends sit-out.
ARES lets JOYADH act before the price drop hits — pull the SKU, raise the floor, prepare the counter — instead of reacting after revenue craters.
ARES observes and recommends. It does not change a single price tag. The final reprice command flows: ARES → JOYADH → (decision) → admin/ops execute. Keeps combat data clean.
| Layer | Tech | Purpose |
|---|---|---|
| Scraper fleet | Playwright · proxy rotation · Bengali OCR | Resilient crawl of 27+ marketplaces + private WhatsApp groups |
| Pricing graph | Time-series db (PostgreSQL TimescaleDB) | Every observed rival price stored with timestamp + URL + raw HTML |
| Combat reasoner | Claude Opus 4.8 + strict-JSON briefs | Turns 200 rows of rival prices into a one-page decision |
| Simulator | Numpy · pandas · margin model | What-if pricing simulations against ONI landed-cost |