JavaScript Tutorial VS ReconPe

ReconPe's answer:

Two architectural choices most reconciliation tools don't make:

• Matching is deterministic, not AI-decided. ACRE (Adaptive Confidence Reconciliation Engine) scores every candidate pair against human-defined rules. Above 90 auto-match, below 40 auto-reject, the 40–90 band surfaces to a reviewer with full evidence. Same input → same output, every time. Auditable, repeatable, defensible to an external auditor.
• The system has a memory. Every open exception is fingerprinted by counterparty, amount, and direction. New exceptions cross-reference the prior 180 days as "Link & Close Both" candidates. Counterparty timing patterns ("late settler, avg lag 5 days") surface after 3 confirmed settlements. The next close starts smarter than the last one. Most reconciliation tools either let an LLM make the match decision (which breaks the audit trail) or treat each month as an island (institutional knowledge lives in analysts' heads). ReconPe does neither.

ReconPe's answer:

• Versus BlackLine and FloQast (enterprise-priced, configuration-heavy, often require professional services to implement): ReconPe is self-serve and starts with a free tier.

• Versus Tally and Zoho Books (general accounting tools with basic reconciliation as a feature): ReconPe is purpose-built for reconciliation — deterministic matching engine, cross-run exception memory, deep marketplace audit logic.

• Versus AI-first reconciliation tools that let an LLM decide matches: ReconPe keeps the match decision deterministic. AI helps with schema mapping, exception investigation, and pattern surfacing — not the match itself. The audit trail stays intact.

• Specific advantage for Indian operators: deep audit on Indian marketplace settlement formats (Amazon Settlement V2 with GST-aware commission, Flipkart's four-fee structure, Meesho price-base, COD aging) that international competitors don't cover well.

ReconPe's answer:

Finance teams running month-end close — controllers, chartered accountants, finance operations leads, and accounting managers.

The natural wedge audience is companies selling on Indian marketplaces (Amazon India, Flipkart, Meesho), where the marketplace audit depth is most differentiated. The same product covers the full GL close — AR, AP, Bank, Intercompany, Fixed Assets, and Inventory — for any finance team running these workflows.

Best fit: mid-market and SMB companies with a finance team of 1–10 people who need reconciliation done correctly but don't have enterprise budget for BlackLine. Also relevant for CA firms and outsourced accounting practices serving multiple clients.

ReconPe's answer:

Started in late 2025 trying to use AI to automate marketplace settlement reconciliation. The early prototype let a language model decide matches. It demoed well. Then real settlement data showed up — Amazon Settlement V2 with GST-aware commission, Flipkart's four-fee structure, exceptions that crossed periods — and the LLM-driven matching produced confident but wrong results that would have misstated ledgers.

That led to the architectural rewrite: a deterministic matching engine in the middle, AI on the periphery for the parts where probabilistic judgment genuinely helps (schema mapping, investigation, pattern detection). Once that was working, we noticed analysts were investigating the same exception against the same counterparty every month, so cross-run exception memory got built.

ReconPe is the result of building, watching it break, and rebuilding with discipline about where AI belongs and where it doesn't. Built solo — code, infra, AI agents, marketing — one keyboard, no handoffs.

ReconPe's answer:

Backend: Java 21, Spring Boot 3.5.10 across 10 microservices (Netflix Eureka registry, OpenFeign clients, Spring Cloud Gateway). Spring AI 1.1.2 with multi-provider LLM support (Anthropic Claude, OpenAI, DeepSeek). PostgreSQL with PGVector for the stateful memory layer. RabbitMQ for asynchronous events between services.

AI orchestration: ReAct planner for the conversational Ask Agent mode, fixed-pipeline orchestration for Investigate Mode. Seven @Tool-annotated Spring beans wired to the chat client.

Frontend: React with TypeScript, Vite, Zustand for UI state, TanStack Query for server state.

Infrastructure: Deployed on AWS — ECS Fargate, RDS PostgreSQL with PGVector extension, S3, CloudFront, Route 53.

Other: JWT authentication via jjwt, Stripe + Razorpay for billing.