Recommended Free Tools
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Kaushal Chaudhari describes a fraud-investigation workflow that uses TigerGraph to assemble transaction evidence, Jev to make bounded choices about further investigation, Bayesian scoring to update risk, and deterministic rules to select actions. A language model explains the decision; it does not decide whether to block a card. The reported “590K+ transactions” is the project’s stated scale, not an independently verified benchmark result.
What the fraud agent is designed to do
The project is an investigation workflow, not an autonomous fraud-blocking system. A case can begin with a customer dispute, a high bank risk score, or an analyst request. The system then gathers a standard evidence pack from TigerGraph before Jev is asked to reason about the case.
Chaudhari describes the division of responsibility this way: “Jev decides when to look. TigerGraph decides what is true.” That is the author’s framing of this implementation, not a general guarantee about graph databases or language models.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteHow the evidence is gathered
A fixed first pass
The initial TigerGraph lookups are the same for every case, according to the project account. This prevents the model from selecting which evidence categories to skip. The graph is named HHGOA; entities include Transaction, BankCard, DeviceProfile, ClosedCase, IdentityFlag, PolicyNote, and ExamCase. Relationships include TXN_ON_CARD, FROM_DEVICE, and CASE_ON_CARD.
#1 Best Overall
| Evidence area | Named lookups | What it contributes |
|---|---|---|
| Transaction and card history | txn_and_card, card_window |
The current transaction and the card’s recent activity. |
| Device and identity | device_profile, device_neighbors, identity_flag |
Device details, connected cards, and identity signals. |
| Region and recurring activity | region_history, recurring_match |
Regional patterns and possible recurring-payment matches. |
| Prior investigations | prior_cases |
Earlier cases connected to the card or related records. |
| Current incident exposure | exposure_episode |
Transactions and exposure associated with the current episode. |
One bounded follow-up lookup
After the fixed retrieval, Jev may choose one additional lookup, such as connected component cards, a bounded card-community walk, prior cases, or a relevant policy passage. Queries were installed in advance and exposed through TigerGraph MCP; the project does not describe generating GSQL at runtime.
Graph retrieval and vector retrieval serve different purposes in this design. The graph establishes which records are connected. Vector search locates a relevant policy paragraph that can support the explanation.
How the score and action are separated
The project uses the bank’s risk_score as a Bayesian prior and updates log-odds using evidence with fixed weights. The score is an estimate used in the investigation; a separate deterministic policy engine chooses the action. A prior case can inform the investigation, but does not alone establish that the current transaction is fraud.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteThe described decision order is:
- Close the case as legitimate when the rules support that outcome.
- Treat an
undocumentedpattern as fraud. - Otherwise, compare the probability with the fraud threshold.
- If the threshold is not crossed, leave the case uncertain rather than forcing a binary result.
A named pattern, a high bank score, or a customer dispute does not by itself trigger every action. Jev can identify a pattern that fixed checks did not surface, request its one allowed follow-up lookup, and check whether explanation sentences are supported by evidence references. It does not select whether to block the card.
Rank #3
What the two example cases show
HHG-011: unresolved evidence leaves the card open
The example begins with the customer statement, “I never made this $131.30 purchase.” The author says the transaction matches card_not_present_new_device. A new device and a connection to a previously confirmed fraud case raise the estimated probability, while a long quiet history lowers it.
Chaudhari reports the illustrative probability moving from around 0.39 to 0.56, then 0.83, and finally back to around 0.76 as evidence is considered. The final result remains uncertain: the workflow opens a case, monitors other cards on the device, and escalates to an analyst while leaving the card open. These are author-reported example values, not independently verified measurements.
Rank #4
HHG-016: assumed customer confirmation changes the plan
For HHG-016, the initial plan is to verify the customer and open a case. In the described scenario, an assumed response confirms the purchase came from the customer’s new phone. The estimated probability then falls to around 0.12, and a second pass through the policy engine closes the case as legitimate. The original and revised plans are retained; alternative replies are stored as counterfactuals rather than presented as actual customer events.
Where human approval and live systems enter
The project describes approval routes for DECLINE_TRANSACTION, FILE_REPORT, and BLOCK_CARD. An approval request is rejected if the action is not on the final action list. Opening a case, monitoring, warning a customer, or closing a transaction as legitimate may proceed automatically in the described design.
Best Value
The benchmark stops short of production execution: it does not send a live SMS, freeze a real card, or write to a production CRM. Consequently, its examples show how plans and approvals are represented, not proof that the workflow has operated against live customers or bank systems.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the reported scale does—and does not—establish
The title reports that the agent can trace a transaction across “590K+ transactions.” Chaudhari’s project article, published on DEV Community on September 25 (the year is not stated on the page), is the basis for that scale claim. The article does not independently establish dataset provenance, benchmark methodology, or fraud-detection performance. The number should be read as the author’s reported project scale, not an accuracy result or evidence of effectiveness in production.
The account also identifies limitations in the current design. Its probability model uses the bank risk score as a prior and fixed log-odds weights. The author says those mappings should be fitted using earlier closed cases and the configuration frozen; that is a proposed improvement, not a reported completed step. The author also proposes an offline community-detection pass and notes that Jev did not always return a probability map in the scored run.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →The project account provides no independently validated performance results, head-to-head model comparison, or sufficient dataset documentation to support claims about real-world fraud detection accuracy. Readers assessing a similar system should look for documented data provenance, calibrated scoring, performance evaluation, and a clear production approval boundary.
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.
