Search
Free Trial

Hongke's latest articles

HongKe

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.

Redis + Graph Database: The best division of labor: real-time scoring to Redis, correlation analysis to Graph?

Many banks are torn between the need to be fast in the moment and the need to look deeper in the aftermath of a transaction when doing AML and fraud. redis and graph databases aren't really one or the other, but rather the most natural combination of tasks: the former is for real-time scoring and the latter is for correlation insights. hkma in its aml/cft regtech guidelines clearly states that banks need to Integrate multi-source data and network analytics To identify behavioral patterns and correlation risks, these two needs correspond to very different technical characteristics - millisecond responses for transactional decisions and multi-hop correlation queries for investigative analysis.

01. Introduction: The two speeds of counterfeiting and AML determine the technical division of labor

There are two distinct time scales for counter-fraud and AML decisions:
  • Transaction Decision Making (milliseconds): FPS RTGS, virtual banking API authorization, digital wallet top-ups - these are all scenarios where risk judgment must be completed before the transaction is released, or the most accurate model will not be able to intercept it. HKMA's TM-E-1 guidelines require banks to implement a risk-based approach to high-risk transactions. real-time monitoringThe customer will be notified of the suspicious activity as soon as it occurs.
  • Survey Analysis (minute to hour level): After a suspicious transaction is detected, it is necessary to track account relationships, fund flows, device networks, merchant relationships and group patterns, and this kind of multi-hops relationship query often takes tens of seconds or even minutes in traditional RDBMS.
Neither Redis nor Graph alone is complete: Redis specializes in high-consolidation real-time reads and writes, but is not suitable for complex graph traversal; Graph databases specialize in correlation analysis, but the query latency is not sufficient for transactional decision-making. The optimal division of labor is Redis for real-time scoring and popular features layer, ArangoDB for association mapping and partnership analysis.The two complement each other through event streams or APIs to form a complete RegTech architecture.

02. Three Core Values: How Redis + ArangoDB Built a Complete RegTech Defense

Value 1: Redis takes on pre-trade decision making, millisecond risk scoring

Pain Points: In an FPS real-time transfer scenario, where the transaction authorization time window is only a few hundred milliseconds, any data query that exceeds 100ms becomes a bottleneck; traditional database queries are more likely to time out during peak hours, resulting in degradation of the release.
Redis response: Redis as online scoring layerIn addition to the above, the system also takes on recent transaction characteristics, device risk, black and white list comparison, velocity metrics and real-time customer risk scores; maintains transaction counts within the time window with Sorted Sets, black list comparison with Cuckoo Filters, and stores session context with Hash, and the response time of all operations is less than 1.5 seconds. sub-1msThe
ArangoDB's division of labor: When the Redis score triggers the "medium-high risk" threshold, transaction events are written to the event stream, triggering ArangoDB's correlation analysis process; Redis is not tasked with in-depth investigations and remains focused on real-time decision making.
The effectiveness of the campaign: The official Redis case study shows that credit card organizations around the world are backed by Redis every second. 700,000 Real-time scoring of transactions, inference speed enhancement 60 timesThe company is also committed to ensuring that scores are always ahead of trade releases.

Value 2: ArangoDB for deep correlation, Redis for caching heatmap results

Response: Purchasing Checklist (8 items) Each of the items in the table below corresponds to "what the board will ask" and "what the HKMA/Internal Controls will look for," which you can take directly to the RFP and score sheet.
Pain Points: AML investigations often need to answer the question of "how many associates are behind this account, where did the funds go, and which merchants did they interact with?" Traditional SQL requires complex JOIN or recursive queries with query times ranging from seconds to minutes, which is not able to satisfy the timeliness requirements of regulatory reports or account freezes.
ArangoDB response: ArangoDB's AQL (ArangoDB Query Language) graph traversal enables multi-hop correlation analysis in a single query, e.g., "Find all associated accounts, devices, IPs, merchants, and money flows within 3 hops of a suspicious account"; ArangoDB Official Fraud Detection Example The official ArangoDB fraud detection case shows that graph ML can identify patterns of collusion that cannot be detected by traditional rules.
Redis division of labor: ArangoDB analyzes the "high-risk association groups" and "hot graph query results" and caches them to Redis, so that the predicted association risk scores can be read directly when making real-time decisions, avoiding the need to repeatedly execute graph queries during peak trading hours.
The effectiveness of the campaign: The application of ArangoDB in financial fraud detection can reduce the correlation query time from Compression from tens of seconds to hundreds of millisecondsIt also improves the accuracy of the model in detecting new forms of money laundering behavior through graph embeddings.

Value #3: Add Decisions platform to automate high-risk investigation processes

Pain Points Open Source may seem cheap, but the bank's "cheapness" is often eaten up by three things: shift labor, incident time, and compliance evidence costs. Conversely, Enterprise may seem expensive, but if it takes your SEV1 recovery time from hours to minutes a few times a year, reduces your upgrade failure rate, and automates your audit evidence, the TCO may be even lower.
 
Response: 3-year TCO Demonstration Model (can be changed directly to your numbers) The following are "sample scenarios" that you can replace with your host count, data volume, region count, and SLO goals:
  • Scope: 6 production clusters (with different business domains), each with 3 masters and 3 slaves across 2 AZs; 2 major version upgrades and 12 minor version/configuration changes per year.
  • Team: 2 SREs (shift), 1 Platform Engineer, 0.5 Safety/Compliance Support (input ratio).
  • Incident: 2 SEV1s per year (with spike jitter/failover/upgrade rollback) with an average of 6 hours of cross-departmental input each time (conservative estimate).
Pain Points: Even with good scoring and correlation analysis, the last mile - the "follow-up of high-risk transactions" - is often stuck in a bottleneck of manual reviews: notifying the customer, freezing the account, filing a SAR (Suspicious Activity Report), and launching an internal investigation, each requiring cross-departmental coordination. Each step of the process requires cross-departmental coordination, and delays can easily result in regulatory fines or capital outflows.
The response of the complete framework:
  • Redis: Score immediately before trading → Mark "High Risk".
  • ArangoDB: Trigger correlation analysis → Generate survey maps
  • DecisionsAutomatic scheduling of follow-up processes, including customer notifications (SMS/Email), account freezes, SAR auto-generation, and survey assignments to AML teams.
The effectiveness of the campaign: This automated process allows high-risk transactions to be minimized. Compressed end-to-end processing time from hours to minutesIt meets the HKMA's requirements for "real-time monitoring and rapid response" while significantly reducing labor costs and operational risks.

Customer Testimonial: RegTech Transformation at an Asian Bank

Background: A large commercial bank in Asia, facing peak FPS traffic (tens of thousands of transactions per second) and increasingly stringent AML regulatory pressure, the traditional risk control system was experiencing bottlenecks in both real-time decision-making and in-depth investigation.
Challenge: During peak trading hours, the fraud scoring latency often exceeded 200ms, resulting in some high-risk transactions being downgraded and released; the AML team's investigation was inefficient, as it took several hours to analyze the association of a suspicious account, and it was even more time-consuming to manually write a SAR report.
Redis + ArangoDB + Decisions driven transformation:
  1. Phase 1 (Redis Instant Layer): Centralize the hot features and risk scores required for pre-transaction decision making into Redis to ensure that FPS transactions are sub-100ms scored before authorization; blacklists and velocity checks are all in-memory.
  2. Phase 2 (ArangoDB Associative Layer): High-risk transactions trigger ArangoDB's graph analytics, automatically generating reports on account networks, money flow mapping, and group identification; AQL queries compress multi-hop correlation times from tens of seconds to hundreds of milliseconds.
  3. Stage 3 (Decisions process level): Automate follow-up with the Decisions platform: real-time customer notifications, account freezes, automatic SAR generation and assignment of investigative tasks; the entire process takes less than 5 minutes from trigger to completion.
Transformation results: Pre-transaction Intercept Rate Increase 35%AML investigates efficiency gains 4 timesThe potential loss and penalty costs avoided each year are more than $10,000 per year. Tens of millions of dollars.The auditability of the framework is also fully compliant with HKMA Regtech requirements.

Conclusion: Act Now to Launch the Blueprint Presentation

As HKMA continues to promote AML/CFT Regtech and network analytics, it is no longer possible to rely on a rules engine or a single database to meet the dual demands of real-time decision making and deep correlation. the Redis + ArangoDB + Decisions division of labor architecture is not a theoretical concept, but a field-tested and proven RegTech. It is a tried and tested RegTech blueprint:Redis guarantees that transactions will not be missed, ArangoDB guarantees that investigations will not be missed, and Decisions guarantees that the process will not be interrupted.The
You have only two choices:
  1. Passive waiting: Continue to use traditional structures to meet increasingly stringent regulatory requirements and wait until the next AML trial or fraud comes to light to remedy the situation.
  2. Active Defense: Let's take a position. Blueprint PresentationIn addition, the program spends 60 minutes answering three key questions:
    1. How many milliseconds is your pre-trade risk rating delay?
    2. How long does it take to analyze the association of a suspicious account?
    3. Has the post-processing of high-risk transactions been automated?
Go to the Redis Product Page

Other Articles
Hongke Case

Rainbow Solutions] Ali's Thousand Questions: How Redis Stream can tame tens of millions of streams in milliseconds?

AIGC has spawned a huge amount of information and news, and it is difficult for traditional message queues to support the high concurrency of tens of millions of information streams. This article explains how to combine Bloom Filter, Consumer Groups, and Vector Retrieval with Redis Stream to realize milliseconds content de-weighting, priority distribution, and real-time recommendation, which can solve the problems of feed stream delay and content homogenization.

Read more
HongKeTechnology May 28, 2026
Hongke Dry Goods

Rainbow Solutions] 2026 Procurement Committee Must-Have: Hong Kong Bank Redis Procurement Checklist (Open Source vs Enterprise Decision Framework)

The HKMA has repeatedly emphasized risk-based and principle-driven requirements in recent years, and has brought the risk of third-party IT solutions to the forefront, meaning that it's fine to turn on Maintenance or Redis Enterprise, but you have to prove that the "controls work" rather than that the "tool is famous".

Read more
HongKeTechnology April 27, 2026
Hongke Dry Goods

Hongke AutoGNSS: How to cross the technical barrier of GNSS complex scene through automated testing?

With the tightening of in-vehicle terminals and intelligent transportation standards (e.g. GB/T 45086), it is difficult to cope with manual testing. HONGKE experts deeply analyze how AutoGNSS automated test platform can realize full chain traceability and data traceability, speeding up the whole cycle of research and development by 4-5 times, and completely eliminating the technological threshold of complex GNSS simulation tests!

Read more
HongKeTechnology May 26, 2026
close menu icon

Contact Hongke to help you solve your problems.

Let's have a chat