Lalithkishore C

Contribute to Deep SWE, building and validating software-engineering benchmark tasks from real issues in open-source repositories. Each task ships with hidden tests and a reference solution, and I set up Dockerized environments that prove the tests fail on the original code and pass once the reference fix is applied.

Contributed fixes to Kestra, the open-source orchestration platform, across the core UI and the serialization plugin. Each fix closed a user-reported issue and shipped after review by the Kestra maintainers — two as my own merged PRs and one co-authored into a maintainer's PR.

• kestra-io/kestra #18071 — Monospace font scale for logs and editorGave logs and the code editor their own monospace size scale (11 / 12 / 14px), so the default medium mode renders them at 12px instead of an oversized 14px, while keeping users' saved font preferences intact. Fixes kestra#17963.
• kestra-io/plugin-serdes #352 — CsvToIon BOM and empty-header fixStripped the UTF-8 BOM that leaked into the first CSV header name, and added an onEmptyHeader option that drops nameless trailing columns, which had broken downstream IonToParquet conversion. Fixes kestra#17646.
• kestra-io/kestra #17640 — Dashboard docs panel fix (co-authored, shipped in #17648)Traced why the No-Code dashboard editor's documentation panel never switched to the selected chart's docs: the editor passed the block's 'type' field where the docs store expected 'cls', so it always fell back to the generic intro. A maintainer folded the fix into kestra#17648 alongside a related fix and merged it with me credited as co-author. Fixes kestra#17579.

Developed backend features and REST APIs with Node.js and Express.js, and worked with the engineering team on building and testing project modules.

A production management system running on the shop floor at Sakthi Auto Components Ltd. Every part gets a QR code, and quality inspectors scan it to log results, so defects and test records are captured in real time instead of on paper registers.

Upload PDFs, Word files, Markdown or JSON and search them by meaning, not just keywords. Every model — OCR, embeddings, reranking, summarisation and classification — runs on your own machine, so no document ever leaves it.

Users see an ordinary drive with folders and drag-and-drop. Behind the scenes a model predicts which files will be needed soon, keeps those fast, and quietly compresses idle files into cold storage — restoring them transparently when they're opened.

Signal pulls official release notes and engineering blogs into one clean feed, refreshed daily, so you can keep up with AI models, developer tools and the cloud without doom-scrolling.

Built for students who learn from YouTube: paste a video or channel and watch it without ads, with a notes panel beside the video and quick access to popular programming channels.

A web app that encrypts and decrypts both images and text, built to make applied cryptography concepts tangible. It beat 50+ teams to win first prize at a state-level project expo.

Took first place among all competing teams with a YouTube Ad Remover that adds an integrated note-taking panel, so students can watch educational content distraction-free and keep notes alongside the video.

Won first prize out of 50+ competing teams for an Image and Text Encryption & Decryption System that demonstrates applied cryptographic concepts on both image and text data.