What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
SentinelGraph, a fraud-investigation prototype described by its builder Nirmal Joseph Ukken, treats a risk score as a reason to investigate—not a verdict. Its agent combines graph evidence, prior cases and policy, then stops when evidence is strong enough, asks for more when it is not, or routes a decision for human approval. The project was published on September 24, 2026, in connection with Task 4 of the TigerGraph problem statement for Hacker House Goa 2026. Its implementation and benchmark results are the builder’s report, not independent validation or evidence of live-bank performance. Read the project post on DEV Community.
What SentinelGraph is designed to do
The central question is familiar to fraud teams: when a risk score pings, is the transaction fraud, a holiday purchase or a new phone? SentinelGraph’s answer is to treat the score as an investigation trigger. As Ukken puts it, “A risk score is a reason to look. Never a verdict.” The system is designed to produce an auditable recommendation and next step rather than an unconstrained language-model verdict.
As an Amazon Associate I earn from qualifying purchases.
Its six-part investigation loop opens a case and audit trail, queries a graph, retrieves similar cases and policy, combines evidence while accounting for correlated signals, decides whether to stop or seek more evidence, and then takes or routes an action while writing the case back to graph memory.
Free tools Windows power users keep installed
One-click scans. No signup required.
How graph evidence and case memory work together
Connecting entities and transactions
The evidence graph links customers, cards, transactions, devices, email domains and billing regions. The project describes 16 installed GSQL queries exposed through TigerGraph MCP. Examples include finding transactions on other cards that share a device or email within a time window, building cardholder behavior profiles, identifying connected card components and retrieving earlier cases through graph adjacency.
#1 Best Overall
Retrieving prior cases and policy
SentinelGraph also uses TigerGraph native vector search to retrieve case and policy memory. The graph traversals and retrieved history provide context for the current alert; the design explicitly seeks to avoid treating multiple correlated signals as independent confirmation. Closed and active cases and policy knowledge are stored in separate graph layers, according to the project description.
When the agent stops—and when it asks for more
The described stop rule requires two independent evidence families to agree before concluding. If they do, the agent stops at a posterior probability of at least 0.85 for the upper threshold or at most 0.15 for the lower threshold. If evidence does not meet either condition, it can request information that may resolve the uncertainty, such as step-up authentication or customer verification. “Uncertain” is an honest answer, Ukken says.
Action depends on policy, not just the probability estimate. The post says some actions are automatic, while L1 or L2 actions wait for human approval. This makes the stopping rule part of a broader process: the system can conclude, gather more evidence or hand off a decision rather than forcing every alert into a fraud-or-legitimate label.
What the language model does—and does not decide
In the described design, deterministic evidence calculations and policy rules govern decision logic. The language model is bounded to extra tool calls and drafting narrative or suspicious activity report (SAR) material, with output validation. This division is intended to keep the recommendation traceable to evidence and policy rather than letting a free-form model response become the final decision.
Rank #3
- Commemorate Tiger Woods' 25-year journey with a billiant, fully illustrated table book from Sports Illustrated
- Sturdy build and construction. The hand bounded green leather hardcover gives it the perfect vintage look and durability
- Its polished aesthetic perfectly aligns with the golf theme of this book, lending an elegant touch to your bookshelf or coffee table.
- 232 pages full of iconic vibrant photos and some of the best written coverage of Woods’s career
- Beautiful Stories, a good read, and great photographies, the ideal gift book for any Tiger fan
What the reported benchmark shows
Ukken says the project used IEEE-CIS card data with the fraud label removed, historical closed investigations, a policy, documented patterns and 20 benchmark alerts. The post reports the following figures; they are the author’s results and have not been independently verified by the cited account.
| Reported measure | Project result |
|---|---|
| Transactions | 590,742 |
| Closed cases | 5,565 |
| Installed GSQL queries | 16 |
| Largest detected ring | 28 cards |
| Memory model AUC | 0.914 |
| Bank score AUC | 0.866 |
| Benchmark alerts | 20: 10 legitimate, 9 fraud and 1 uncertain |
| SARs in the benchmark | 6 |
The post says the model was trained on July through September and tested on October, and that rerunning the 20 alerts produced the same decisions. It also says a gradient-boosted model trained on closed cases achieved nearly double the bank score’s average precision, but gives no exact average-precision values. These are project-reported results on a defined benchmark; they do not establish performance on live transactions, generalization to other data or operational savings.
What remains outside the demonstrated system
The author says evidence replies were simulated because real reply channels had not been implemented. Real SMS or app replies, streaming ingestion, likelihood ratios learned from resolved agent cases rather than set by hand, and external enrichment are listed as future improvements. The post therefore describes a prototype and benchmark, not a fully live end-to-end banking service. Its reported repeatability applies to the 20-alert benchmark, not necessarily to other cases or deployments.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteHow to assess the project as a fraud-team design
For teams evaluating an agentic investigation workflow, SentinelGraph raises useful questions beyond model accuracy:
- Evidence traceability: Can a reviewer follow each finding back to graph relationships, prior cases and policy?
- Correlated signals: Does the system distinguish genuinely independent evidence from several indicators that reflect the same underlying fact?
- Stopping and uncertainty: Are thresholds explicit, and can the agent defer a conclusion when evidence is insufficient?
- Human approval: Which actions may execute automatically, and which require a reviewer?
- Case memory: How are resolved cases retrieved, and how are their lessons incorporated without mistaking similarity for proof?
- Operational readiness: Are replies, streaming ingestion and external enrichment actually connected, or simulated or planned?
- Evaluation quality: Are results limited to a project benchmark, or independently replicated on representative live operations?
The project provides a concrete example of combining graph traversal, case memory and explicit uncertainty behavior. Its reported AUC comparison and repeatable benchmark are evidence about the author’s test setup, not a vendor comparison or proof that the approach will perform similarly in a bank’s production environment.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
