Agastya RAG
Agastya RAG is a training-free, structure-aware retrieval system that helps AI agents navigate complex documents efficiently. Runs on a single 8GB GPU, and supports auditable retrieval without fine-tuning.
The bigger builds: AI systems, product work, and hardware that took more than an afternoon. Smaller experiments live on the Experiments page.
Agastya RAG is a training-free, structure-aware retrieval system that helps AI agents navigate complex documents efficiently. Runs on a single 8GB GPU, and supports auditable retrieval without fine-tuning.
An intelligent research assistant for academic libraries, exploring how far LLMs can go in making research easier to navigate. Pitched at the Bardusch Family IdeaMakers Challenge, 2026 (2nd place).
Weather along the drive, at the time you will actually be there. The prototype samples the route geometry instead of just the two endpoints, asks for the forecast at each waypoint's real arrival hour, and has an LLM write the briefing a driver wants before setting out.
Type a theorem, get a typeset proof back as a PDF. Two model calls in sequence: one writes the proof, the second only reformats it as LaTeX, then pdflatex compiles it.
A Python virtual assistant built on OpenAI's API: context, tools, and real tasks rather than canned commands. An early, deliberate take on what an assistant could become.
An IoT system designed for classrooms to save money on electricity, built and presented at the CBSE Science Fair 2022 with a working demo on video.
An early applied-ML project: classifying stars from data with a trained model; one of the builds that pointed everything toward AI.
A rover built from the ground up and documented in a ten-minute walkthrough: wiring, drive, and control in one build.
A full automation pass over my own room: a four-channel relay board and an HC-05 Bluetooth module putting lights and fan under phone control; lived with daily for months while the bugs got worked out.
The automated garden, evolved into a self-regulating mini greenhouse: soil, rain, and light sensors decide when to water. Presented at the CBSE Science Fair 2019, with the full build on video.