AAKASH DAHIYA
>
I build and ship LLM-powered products end to end — retrieval-augmented generation, multimodal vision pipelines, and the Python/FastAPI and Next.js/Supabase systems around them. MSc in Computer Science. Currently the sole developer of a live two-app platform for a used-appliance business in Delhi NCR.
README.TXT
I'm an AI engineer with an MSc in Computer Science from Lakehead University, and I like the parts of this job that other people find annoying: making a retrieval pipeline actually return the right chunk, getting a vision model to produce copy a human would sign off on, and wiring two apps together so a business owner can move an item from "just bought" to "live on the store" in one click.
At CashFlip I'm the only developer. That means I own the Next.js frontend, the Supabase schema, the Claude vision integration, the WhatsApp order flow, the admin panel, and the Google Ads attribution — and I get to see exactly which decisions moved the number.
Outside of work I build things like TubeQuery, a RAG app that answers questions about any YouTube video, mostly because I wanted to know what breaks when you cache embeddings badly. (A float32/float64 dtype mismatch. It silently corrupts every similarity score.)
EXPERIENCE.EXE
Software Developer · CashFlip
Delhi NCR · Remote · Sole developer- +35%Increased lead conversion by 35% by replacing a manual, call-based quoting process with an online quote wizard and photo uploads.
- Built a lead-generation web app (Next.js, Supabase) that gives sellers instant quotes for used appliances.
- Built the CashFlip Store (shop.cashflip.in) with an AI listing generator that uses Claude vision to turn item photos into product titles and descriptions.
- Connected both apps through a shared Supabase Postgres backend, with a one-click "List on store" flow that moves purchased items from the buying pipeline to the storefront.
- Implemented cash-on-delivery checkout with WhatsApp order confirmations, plus an admin panel with a pricing editor.
- Built keyword-level lead attribution on top of Google Ads conversion tracking, so the admin dashboard shows the search term behind each quote.
PROJECTS.DIR
TubeQuery
RAG Q&A over YouTube videos
- Full-stack RAG app that chunks YouTube transcripts, embeds them with all-MiniLM-L6-v2, retrieves by cosine similarity, and generates grounded answers with Claude Haiku 4.5.
- Designed a FastAPI backend with an embedding cache and a vanilla HTML/JS frontend.
- Debugged a float32/float64 dtype mismatch that silently corrupted every similarity score after caching.
CashFlip Store
AI listing generator + storefront
- Storefront where Claude vision turns raw item photos into product titles and descriptions, so listings go up in seconds instead of minutes.
- Cash-on-delivery checkout with automated WhatsApp order confirmations.
- Shares one Supabase Postgres backend with the quoting app — one click promotes a purchased item into a live listing.
SKILLS.CFG
LANGUAGES
AI / ML
BACKEND & DATA
FRONTEND
TOOLS
EDUCATION.LOG
MSc, Computer Science (course-based)
Lakehead University — Thunder Bay, ON
Relevant coursework: Pattern Recognition
2026Bachelor of Computer Applications (BCA)
Bharati Vidyapeeth Institute of Management & Research (GGSIPU) — New Delhi
2025LET'S BUILD SOMETHING
I'm open to AI / Python engineering roles in the Toronto area or remote. If you're shipping something with LLMs, I'd like to hear about it.