I build speech AI, LLM evaluation and the full-stack products around them, and once taught a robot to find its own way.
Faisal Wani
You dream it, I code it to life. I’m Faisal 
, an engineer who builds speech AI, LLM tooling and the full-stack products around them. I studied Digital Engineering at PSU, built a robot 
that maps its own room, and when I’m off work I’m on a football pitch 
.
- Research Engineer at SCB DataX
- Iqra AI, a voice tutor for Quran recitation
- SCB 10X, Mojen.ai, SCG and Booktree
- B.Eng Digital Engineering, Prince of Songkla University
Where I’ve worked.
Real-time Thai speech recognition, and the evaluation and monitoring that take a bank’s voice agent to production.
- FastAPI
- DeepEval
- Langfuse
- OpenTelemetry
- Azure AKS
- Co-authored the Typhoon ASR Streaming paper on real-time Thai speech recognition, with 4.5× fewer errors than the full-context baseline, and built the FastAPI and WebSocket service that serves the model.
- Built the evaluation harness for a bank’s five-service voice AI agent: it scores real conversations end to end with LLM-as-a-Judge (DeepEval), and other teams run their own test sets through its API.
- Instrumented the agent with OpenTelemetry across Python, Go and Java services, with a Langfuse and ClickHouse dashboard for quality, latency, cost and safety on live traffic.
- Hardened it for the bank’s security review (Entra ID single sign-on, group-based access, managed identity instead of API keys, PII access logging) and shipped it to AKS through the bank’s Jenkins and Helm pipeline.
- Turned the Linguist team’s annotation platform into a multi-project speech-data pipeline and used it to prepare a Thai ASR benchmark.
Things I’ve built.
Iqra AI
A real-time voice tutor for Quran recitation. It listens while you recite, grades every sound, and tells you where in the mouth or throat a missed sound is made.
- phoneme error rate, under the 9.5% published on the same data
- 8.9%
- lessons, from the Arabic alphabet to tajweed
- 66
- LiveKit
- wav2vec 2.0
- Modal
- Next.js
Alwani Travel
The booking platform of a licensed Phuket travel agency: tours, tickets and transfers from search to checkout, rebuilt in 2026 on Next.js, NestJS and PostgreSQL.
- tours, tickets and transfers on sale
- ~80
- tasks shipped in seven months through a multi-agent workflow
- 350
- Next.js
- NestJS
- PostgreSQL
- Claude API
Alwani DMC
The trade side of Alwani Travel: a bilingual site where travel agents and tour operators see what the Phuket team runs on the ground, then send a brief for a costed plan.
- destinations, drawn on one map as you scroll
- 5
- languages, English and Thai
- 2
- Next.js
- GSAP
- Lenis
Earlier builds
More on GitHubDragon Ball NFT Game
A memory card game where winning cards are minted as NFTs you can trade on OpenSea.
Autonomous Service Robot
A mobile robot that maps a room with LIDAR and plans its own collision-free path.
Andaman Luxury Car
A Thai and English site for a Phuket chauffeur service with a black Toyota Alphard, told as a scroll film: the car, the island it drives you around, the fares and a form that writes the booking for you.
What I build with.
The languages, frameworks and infrastructure I use most. Tap one to toss it.
Certificates and awards.
Academic
Degree, GPA awards and international GCSEs.
5
Competitions
Hackathons, coding contests and maths olympiads.
5
Courses
Data, AI and event streaming fundamentals.
3
Off the clock.
When I’m not at a keyboard, I’m on a football pitch. A few trophies came home too.
