Hacker

Yash Chandnani

@yash_chandnani

Joined Oct 2026Student at Symbiosis institute of technologyIndia

Projects

  • Voila — GST Close Room on Switch

    Flip the Switch HackathonNov 2026

    Voila solves a specific and high-risk compliance problem for Indian small businesses: the last-mile GST reconciliation before filing, where accountants must verify every purchase invoice against GSTR-2B and decide what Input Tax Credit is truly claimable within a few days. It is built for small business owners, their accountants, and Chartered Accountants who are under time pressure, often working with messy spreadsheets and WhatsApp vendor messages, and need a reliable way to catch mismatches, blocked credits under Section 17(5), and missing supplier filings before they sign off on a return. The product is especially useful for teams that need trustworthy, auditable decisions rather than black-box automation—people who need a system that can explain why a claim is valid or denied, track approvals, and produce a filing-ready GSTR-3B draft with a clear audit trail.

    Switcher@anushka_salvi,@akkki007

  • Kaizen: find the commit that broke production

    Graph HacksOct 2026

    When a deploy breaks production, the hardest question is the first one: which change did it? A single deploy ships many commits. Logs tell you what failed, Git tells you what changed, and nothing connects the two. So the on-call engineer greps logs at 3 AM, guesses, and usually rolls back the latest commit. On our benchmark that guess is wrong 68% of the time. Kaizen connects the two in a FalkorDB knowledge graph. It reads the repository's history and maps every function, which function calls which, which endpoint each one serves and which commit changed which function. Production errors are linked to the exact function that threw them. When something breaks, Kaizen: detects it on its own (error rate over its own baseline for two 30-second windows); walks the graph from the stack trace back to the commits in the latest deploy and ranks them with ten graph features, naming a culprit only when the evidence is clear; tells the team in Slack: the culprit commit and author, the function that changed versus the one that threw, the confidence and a link to the evidence; explains itself: every score shows the Cypher query behind it, and an AI agent writes an explanation that may only cite nodes the graph returned; fixes it safely: proposes a revert, verifies it in staging against the same failing traffic, and merges nothing until a person approves; remembers: the fix and the confirmed cause are written back to the graph, so a similar bug later is recognised faster. Who it's for: engineering teams who ship often and own their production, especially on-call engineers and small teams without a dedicated SRE function. It plugs in with a GitHub App and one ingest key. Does it work? On 28 replayed incidents full of decoy commits, Kaizen ranks the real culprit first 89% of the time, against 39% for keyword search and 32% for "blame the latest commit", with 0% false attribution (latest commit: 68%). It scores 83% on held-out scenarios it was never tuned on. Removing the graph's shortest-path feature drops it to 57%, so the graph is doing the work, not the LLM.

    Keizoku@kamran_03,@sanskarjoshi,@akkki007