projects

whatsapp-first print-ordering demo for indian print shops — customers send PDFs and instructions, Gemini normalizes print details, Cloudflare Workers stores files in R2 and orders in D1, Razorpay Test Mode handles payment links, and a Hono dashboard lets shops process paid orders.

real-time ai meeting copilot — captures microphone audio in segmented chunks, transcribes via Whisper, and surfaces live suggestions (questions, talking points, answers, fact-checks) with a streaming chat panel for detailed follow-ups and context compression for long sessions.

maritime supply-chain intelligence platform — a NetworkX graph world model simulates disruption cascades at global chokepoints, with NLP entity extraction, sentence-transformers RAG, and streaming OpenUI Lang responses rendered as interactive React components on a 3D globe.

voice support app for e-commerce — streams live conversations to Gemini Live, grounds replies through tool-calling over local order and policy data, and responds in both audio and synchronized chat text.

ai-powered content editor for structured marketing drafts — generates social posts, blogs, and landing pages block-by-block with SSE streaming, grounds outputs on uploaded files via semantic retrieval, and stores checkpointed version history in Postgres.

chrome extension that detects manga speech bubbles using computer vision (RT-DETR-v2), translates them with Gemini across 12 languages, and renders live overlays in-page — cached for instant reuse without touching the original artwork.

ai-powered resume screening platform — a LangGraph multi-agent pipeline extracts skills, matches candidates, and returns deterministic match scores via a weighted formula in code. supports batch upload, per-result resume viewing, and human-review flags for borderline matches.