Swastik Agrawal
Electrical & Electronics
Engineer.

A retrieval-augmented generation chatbot that answers questions strictly from a PDF's content. Chunks large documents intelligently, lets users swap PDFs at runtime, and runs smoothly on basic hardware — no expensive GPU needed. Scope is intentionally bounded to the document context, eliminating hallucinations entirely.

A reliable AI chatbot with automatic LLM failover — primary GPT-4 OSS 120B, auto-switches to Meta Llama 3.3 70B on downtime. Stores conversation history in SQLite and sends the last 250 messages as context on every prompt, making answers genuinely coherent over long sessions. Smart prompt limits prevent information overload.

PennyTrack is a personal expense tracking desktop app built with Python. It lets you log and categorise your daily spending, then filter it by today, this week, this month, or all time. Expenses are stored locally using SQLite, keeping your data private and accessible offline. Each transaction supports a category, date, amount, and an optional note.

MailKey is an email OTP authentication system built with Python. It sends one-time passwords via SMTP, optimised to land in the inbox. The sending domain was verified and configured through Cloudflare — setting up the necessary DNS records to establish trust with receiving mail servers. Clean, lightweight, and easy to integrate into any project.

Gravity is a voice-operated desktop assistant built with Python — one of my earliest projects. It uses speech recognition and text-to-speech to interact with the user, handling daily tasks through straightforward command logic. Simple by design, but meaningful in what it taught me about building things that actually respond to the real world.