Most of what I build watches something quietly — services breathing, audio streaming, logs scrolling — and steps in only when it matters.
I’m Lakshya. I’ve spent the last seven years moving between front-end and back-end work, chasing a particular kind of problem: systems that need a careful pair of hands to keep them upright. Heartbeat agents pinging plant infrastructure. Whisper pipelines transcribing audio in chunks. Log analyzers that read their own stack traces.
I’m happiest when the work is quiet, durable, and slightly unglamorous — when a tool you wrote three months ago is still running, and nobody had to think about it.
Built end-to-end for a Manufacturing Execution System — a no-code admin panel to configure chatbots (prompts, model, RAG knobs) and a self-contained widget that drops into any page with one <script> tag, isolated by Shadow DOM.
Engine combines LlamaIndex + Qdrant RAG with a data-agent that runs read-only SQL and internal API calls, returning live plant data as tables and 11 chart types. LLM provider is switchable per-chatbot (Azure OpenAI · OpenRouter · OpenAI). Nginx fronts admin + widget same-origin under the main MES app with fail-safe upstreams.
A Python agent that watches multi-service apps in real time, catches crashes the moment they happen, and walks the trace itself.
When something breaks, an AI pipeline reads the stack, locates the seam, and proposes an immediate code fix — front-end or back-end. Slack and email pings carry the diff, not just the error.
Built as a free alternative to enterprise call assistants — and earned its community traction by refusing to put the useful parts behind a paywall.
It listens to live audio for in-call advice, lets you drop in UI screenshots and returns ready-to-use front-end code, and parses your résumé into 15 specific interview questions tailored to it.
I'm open to contract work, full-time roles, and small open-source collaborations — especially anything sitting between AI, audio, and infrastructure.