FinTruth audits the numbers, facts, and claims behind every machine-written financial document — intercepting hallucinations before they become financial risk. Built by Acellent Technologies, a wholly-owned subsidiary of Advanced Biomed Inc. (NASDAQ: ADVB). FinTruth 对每一份机器生成的金融文件中的数字、事实与论断进行系统性审计——在 AI 幻觉演变为金融风险之前将其拦截。由 Acellent Technologies 研发,该公司为 Advanced Biomed Inc.(NASDAQ: ADVB)的全资子公司。
Fabricated figures, confused entities, contradictory logic — LLMs introduce six categories of hallucinations into financial reports, directly threatening credit, compliance, and regulatory decisions. 虚构数据、主体混淆、内部矛盾——大模型在金融报告中引入六类典型幻觉,直接危及信贷决策、合规审查与监管报送的安全边界。
Fabricated financial figures, calculation errors, inconsistent totals across statements. 虚构财务数据、计算错误、报表间数值不一致。
Confused affiliates, misattributed identities, broken guarantor chains. 关联方混淆、主体身份归属错误、担保链断裂。
Wrong event sequence, incorrect dates, anachronistic references. 事件顺序错误、日期错位、时点引用倒置。
Internal contradictions between sections, reasoning gaps, unsupported leaps. 章节间内部矛盾、推理链跳跃、未支撑的结论。
Incorrect policy citations, regulations misquoted, sources misattributed. 政策法规引用错误、监管文本错引、来源归属偏差。
Overly optimistic or pessimistic framing, sentiment-rating mismatch. 过度乐观或悲观的表述、情感与评级不匹配。
Progressive validation: hard constraints guarantee baseline safety, logic ensures correct reasoning, soft constraints elevate quality. 渐进式验证:硬约束保障安全底线 → 逻辑层确保推理正确 → 软约束提升报告质量。
Must be 100% satisfied. Any violation triggers an error and blocks output. 必须 100% 满足。任一违反即触发错误,阻断输出。
Validates reasoning chains, ensuring conclusions follow from premises and evidence. 验证推理链有效性,确保结论由前提与证据合理推导。
Confidence-scored: violations are tolerable but flagged for review and quality scoring. 置信度评分:违反可容忍但会标记供复核,并参与质量评分。
Each module attacks a different failure mode — numbers, logic, ontology, narrative — then converges into a unified audit trail. 四个引擎各自针对不同的失败模式——数值、逻辑、本体、叙述——最终汇聚为统一的审计链。
FinTruth wraps Smith RM — the financial reasoning core — with a Formal Verification Engine and a financial rules knowledge base, end-to-end auditable. FinTruth 以金融推理内核 Smith RM 为中枢,外层嵌套形式化验证引擎与金融规则知识库,全链路可审计。
Five stages from raw data to narrative report. At each stage, a paired verification checkpoint produces evidence and a confidence delta. 从原始数据到叙述报告的五个阶段,每个阶段都配套一个验证检查点,产出证据与置信度增量。
Credit reports, financial statements, business registry data. 采集信贷报告、财务报表、工商登记信息。
Cross-source validation, missing value detection. 跨数据源交叉验证,缺失值检测。
Smith RM performs financial analysis on aligned data. Smith RM 在对齐后的数据上执行金融分析。
Financial data consistency, industry benchmarking. 财务数据一致性,行业基准对标。
Risk factor identification and severity assessment. 风险因子识别与严重度评估。
Risk-evidence chain completeness, guarantee logic. 风险与证据链的完整性,担保逻辑。
Credit rating generation with full reasoning trace. 生成信用评级并保留完整推理过程。
Rating-indicator consistency, historical reasonability. 评级与指标一致性,与历史评级合理性比对。
Narrative report output with embedded evidence anchors. 生成带证据锚点的叙述性报告。
Narrative-data consistency, sentiment-rating alignment. 叙述与数据一致性,情感与评级对齐。
Claims classified as pass / suspect / error. Click any claim — the report highlights its position, and the source PDF jumps to the exact paragraph. 每条论断被分类为通过 / 存疑 / 错误。点击任一论断,报告自动高亮其位置,源 PDF 同步跳转至对应段落。
The company reported total revenue of ¥1.247B for fiscal year 2025, representing a year-over-year growth of 14.2%. Profitability remained robust with net margin at 18.4%, comfortably within industry benchmarks. 该公司在2025财年实现营业收入 ¥12.47亿,同比增长 14.2%。盈利能力保持稳健,净利率达 18.4%,位于行业基准区间内。
The guarantee structure involves ABC Co. as primary guarantor — however, the registry shows two affiliated entities sharing this name; the report does not disambiguate. 担保结构中,ABC公司作为主要担保方——但工商登记显示存在两家同名关联实体,报告未予区分。
Leverage analysis indicates EBITDA-to-debt ratio of 0.42, suggesting moderate financial flexibility. The credit committee assigned a rating of BBB+ based on these indicators. 杠杆分析显示EBITDA / 负债比为 0.42,表明财务弹性中等。信贷委员会基于此指标给出 BBB+ 评级。
Auto color-coded by status: pass / suspect / error across the entire report. 按状态自动配色:通过 / 存疑 / 错误,贯穿整份报告。
Independent chapter scores: completeness, data accuracy, narrative quality. 章节独立评分:完整性、数据准确性、叙述质量。
Real-time validation time, token usage, retry counts, system health. 实时验证耗时、Token 用量、重试次数、系统健康度。
Export audit reports with conclusions and full evidence chains. 导出审计报告,含结论与完整证据链。
From commercial banks to regulators — wherever AI-generated financial content meets capital, compliance, or credit decisions. 从商业银行到监管机构——凡AI生成的金融内容涉及资本、合规或信贷决策之处,皆为我们服务的场景。
FinTruth is the financial verification product of Acellent Technologies (Hong Kong) Co. Limited, a wholly-owned subsidiary of Advanced Biomed Inc. (NASDAQ: ADVB) since April 2026. Live in production at a top-tier Chinese commercial bank, serving roughly 1,000 credit managers daily. FinTruth 由 Acellent Technologies (Hong Kong) Co. Limited 研发,该公司自 2026 年 4 月起成为 Advanced Biomed Inc.(NASDAQ: ADVB)的全资子公司。产品已在中国某头部商业银行投入生产,日均服务约 1,000 名信贷经理。
Visit Investor Center 访问投资者中心 →Book a live demonstration. We'll audit a real AI-generated report from your team in 15 minutes — and show you exactly what the model got wrong. 预约一场实时演示。我们将在 15 分钟内对您团队的一份真实 AI 报告进行审计,并明确指出模型在哪里出错。