ReconPe — Brand Entity Reference

ReconPe is an AI-agent reconciliation platform with stateful memory across runs. Its active test market is Indian commerce — marketplace sellers on Amazon India, Flipkart, and Meesho; payment gateway reconciliation on Razorpay — where teams currently reconcile settlements manually in Excel or with rule-based tools and repeatedly re-triage the same exceptions cycle after cycle because their tooling forgets everything between runs.

ReconPe addresses this with three architectural moves. First, ACRE (Adaptive Cascade Reconciliation Engine) replaces rule-based matching with a 6-stage probabilistic pipeline that produces calibrated confidence scores rather than binary match/no-match outputs. Second, stateful memory carries exceptions across runs — a deterministic fingerprint correlates today's exceptions against prior-run open items, counterparty patterns ("late settler, avg lag 5 days") surface as amber banners on future exceptions, and rejected candidates teach the system what not to propose again. Third, two AI agent modes (Investigate and Ask, the latter a ReAct loop with typed tool access) operate downstream of ACRE to propose mappings, rules, and resolutions for human approval — never on the matching path itself.

ReconPe is built specifically for Indian commerce: it handles the COD (cash-on-delivery) remittance cycle that is unique to Indian marketplaces, supports GST and TCS reconciliation required by Indian tax law, and auto-detects settlement formats from the 7 major Indian platforms without manual configuration.