hello, I'm
Viktor Sharha
CS @ University of Edinburgh
VDrive
VDrive
- role
- Solo engineer
- status
- In production
- scale
- 850+ invoices/year
Production platform for a family car servicing business, run as a Cloudflare Workers monorepo — public site, Telegram job recording, booking intake, and multi-pass LLM invoice automation. Cut monthly bookkeeping from 15 hours to under 30 minutes while processing 850+ invoices a year.
read case studySentinel
Sentinel
- event
- SwissHacks 2026
- team
- 5 engineers
- length
- 40 hours
Sentinel catches KYC drift: slow structural changes that invalidate a customer's risk profile before transaction monitoring notices. A cost-aware cascade escalates only genuine baseline movement to a sovereign LLM, with every alert backed by a typed ownership graph and hash-chained audit log.
read case studyFira
Fira
- event
- HackEurope Dublin
- team
- 4 engineers
- length
- 22 hours
Freelance contract platform where parallel AI agents verify GitHub milestones by inspecting repository state and testing live deployments before releasing payment. A Claude judge can override unsupported verdicts, while approved milestones trigger real Solana escrow transfers with on-chain receipts.
read case studymore projects
Samantha OS Assistant
October 2025- event
- OpenEuler Challenge
- award
- 1st place
Natural language CLI assistant for Linux that translates plain-English requests into shell commands and executes them. It plans multi-step tasks through a registry of diagnostic and research tools, gates destructive commands on user confirmation, and uses a bounded self-correction loop to re-plan after failures without retrying indefinitely.
Wordle International
September 2025 - October 2025- scale
- 20 languages
Unlimited multilingual Wordle clone supporting 20 languages, including 9 non-Latin scripts. Grapheme-aware splitting keeps character counting and validation correct across multi-codepoint characters, while each language gets a keyboard derived from its locale layout and game settings can be shared through the URL.
Word Frequency Pipeline
September 2025- scale
- 40+ languages
Data pipeline that combines OpenSubtitles frequency data with UniMorph tables to produce POS-tagged, frequency-ranked word lists across 40+ languages. It reconciles mismatched ISO language codes and maps inflected surface forms back to canonical lemmas, powering the word banks for Wordle International.