Developer scores 66/100 after AI judge quizzes him on his AI-built hackathon project

·1 views

A developer entered HackerRank's Orchestrate hackathon in September 2026 and built a financial decision tool called Praxi Clew, which forecasted 90-day balances to assess whether a user could afford an expense. The tool was largely constructed using AI coding assistants, including Antigravity with Gemini and ChatGPT Astra, while the developer directed the work. After submitting the project, he underwent a 30-minute voice interview with HackerRank's AI judge, Chakra, which asked adaptive follow-up questions probing his understanding of the architecture, implementation details, and trade-offs. The developer acknowledged mid-interview that he struggled to explain technical specifics of a system he had built with heavy AI assistance, ultimately scoring 66.1 out of 100. He concluded that being able to articulate how something was built should be an integral part of the building process itself.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer Ships 4 AI Agent Payment APIs on Base Mainnet, Earns Zero Revenue

A developer has published a detailed build log after deploying four AI-agent payment APIs on Base Mainnet, reporting zero external revenue. The project encountered several real-world technical issues, including a facilitator hang and a silent scanner bug that went undetected for some time. Despite the systems functioning on a live blockchain network, no paying external users were acquired. The developer reframes the lack of revenue not as a technical failure but as a market signal, suggesting demand for this type of product may not yet exist. The candid post-mortem offers insights into the gap between functional Web3 infrastructure and actual product-market fit.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer finds AI coding agents routinely ignore their own stored rules

A developer ran controlled tests across hundreds of sessions to measure whether a coding agent actually follows rules loaded into its memory, finding the results far worse than expected. Despite relevant rules being retrieved and displayed to the model, the agent repeated the same heredoc-related coding mistake four times in a single conversation. Tests showed the agent followed a pointer instruction to read a rules file in only 4.9% of 89 sessions, and even with all 59 rules loaded simultaneously, violations still occurred six times in one day. The author found that neither preloading rules nor fetching them on demand could reliably stop a model once it chose to ignore a rule, and that enforcement required hook-based checks at the tool-call or turn-completion stage. The investigation also revealed that rule compliance varied significantly across different agent hosts and that many rules lack observable outputs, making automated enforcement structurally difficult.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Tests Amazon Bedrock AgentCore Harness With a Minimal CloudOps Agent

A developer building a CloudOps agent used Amazon Bedrock AgentCore's managed Harness to run a first experiment with no external tools, no memory, and a restrictive system prompt. When asked whether production was healthy, the agent correctly refused to answer, citing its lack of access to live infrastructure or monitoring data — the boundary the developer intended to verify. The setup revealed two practical lessons: a descriptive Harness name caused CDK synthesis to fail due to a name-length limit, and the AWS environment required bootstrapping before a dry-run deployment could complete. An initial attempt using an Anthropic model also failed because the account lacked approved access, prompting a switch to Amazon Nova Micro. The project is described as Stage 1 of a broader effort to build an agent that reads deployment status by calling a read-only Lambda function.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How the Fed's Policy Type, Not Just Its Speed, Determined the 2020 Market Bottom

During the 2020 COVID crash, the S&P 500 fell 35.3% over 20 days, from February 19 to March 23, as forced selling cascaded through leveraged markets. The Federal Reserve intervened three times, but its first two actions — emergency rate cuts on March 3 and March 15 — targeted the price of money rather than the flow of forced selling, and failed to halt the decline. The March 15 cut of 100 basis points actually accelerated the selloff, triggering a circuit breaker the following day as markets interpreted the move as a sign of deeper distress. Only on March 23, when the Fed announced open-ended purchases directly targeting the assets being liquidated, did the cascade stop — with the market bottoming that same day. The analysis argues that standard risk models misattribute the tail risk by treating Fed intervention as a residual, when the true variable is whether policy binds to price or to selling flow, and how many days elapse before an effective backstop arrives.

0 comments Read more at DEV Community