AI Trust & Safety & Model Quality Specialist MBA • Meta L4 Authority

Abiral Parajuli

Location
Earl's Court, London
Email
aviral.prz@gmail.com
Phone
+44 7436598814

Results-driven AI Trust & Safety and Quality Specialist with an MBA and extensive experience at Meta and enterprise AI pipelines. Proven track record in LLM reliability enhancement, Root Cause Analysis (RCA), adversarial risk mitigation, and vendor calibration across high-stakes policy domains. Adept at bridging technical AI development with operational risk execution, designing scalable evaluation frameworks, and closing model reasoning gaps.

View experience → Impact at a glance

01

Impact at Scale

Verified incident logs • RCA closure

WEIGHTED_RECALL

11.1pp

+11.1 increase

UNWEIGHTED_RECALL

5.4pp

+5.4 increase

TIME_SAVED

10h/week

AUTO tally fix

ESCALATED_REVIEWS

200+

HIGH-STAKES Gambling & Games

CALIBRATION_SESSIONS

30+

QUARTERLY vendor ops

ACCURACY_BOOST

8%

+8% classification

CERT_RATE_LIFT

12%

+12% first-pass

DATA_THROUGHPUT

22K/week

MULTIMODAL QA volume

BENCHMARK_ACC

95-97%

STRICT MAG7 client

DEFECTS_RESOLVED

25

+20% eff tooling

02

Skills & Expertise

AI Safety & Trust & Safety

  • Frontier Risk Mitigation
  • Trust & Safety
  • LLM Quality Assurance
  • Root Cause Analysis (RCA)
  • Adversarial Risk Mitigation
  • Edge Case Resolution
  • Policy Evaluation
  • Vendor Calibration

Data & Analytics

  • SQL
  • Python
  • Pandas
  • Tableau
  • A/B Testing
  • Statistical Data Analysis
  • Data Visualisation

Tools & Formats

  • JSON
  • Label Studio
  • Google Sheets
  • Meta L4 Authority Tools

03

Experience

  • Led RCAs closing policy bypass vectors 11.1pp weighted recall and 5.4pp unweighted
  • Spearheaded cross-functional workflow resolving critical platform tally counter defect saving 10h weekly
  • Partnered with legal and project manager to support development of Meta's new Gambling Scams policy for LLM training
  • Served as cross-functional SME and POC for high-stakes Gambling & Games policies aligning 200+ escalated reviews
  • Directed vendor quality operations through 30+ quarterly calibration sessions hitting certification in 4 days boosting accuracy 8%
  • Designed and launched gamified onboarding framework increasing first-pass certification 12%

Executed QA on up to 22,000 multimodal data points weekly for MAG7 client maintaining 95-97% benchmark

Managed high-volume daily cash desk operations overseeing £50,000+ daily transactions 99.8%+ accuracy

Education & Certifications

University of Westminster

2021-2022 • MBA

UK

Sichuan University

2017-2021 • Bachelor's Civil Engineering

CN

Certifications

  • L4 Authority Competency Calibration (Meta, 2026)
  • Data Analyst Professional Certificate (Meta, 2026)

Languages

  • Nepali • Native100%

  • English • Bilingual100%

  • Hindi • Bilingual100%

  • Chinese • Limited45%

Contact

Let's build safer AI systems together.

  • Earl's Court, London
  • aviral.prz@gmail.com
  • +44 7436598814

Eligible to work in the UK — Partner Visa

04

Playground — Policy Patrol

TRUST & SAFETY SIM

Gamified onboarding prototype from Meta vendor training — jump over risky gambling vectors. Built to teach policy reasoning under speed pressure.

Training simulator — Policy Patrol Ready
READY

P / Esc to pause • Space to jump

Vectors Cleared: 0Best: 0

POLICY PATROL • GAMBLING & GAMES

Ready to patrol?

The shield is your trust & safety defender. Obstacles are risky gambling products: slot machines, playing cards, dice, roulette chips, and SCAM tokens. Jump to keep the platform safe.

🎰 slot 🃏 card 🎲 dice ⭕ chip 🚫 SCAM

Press Space, Up arrow, or Tap / Click to jump over risky gambling products

P / Esc to pause • Space to jump • Space to restart

Space / ↑ to jump • Tap canvas to jump • Space restarts after game over

Tap to jump • Use top bar to pause/resume • Tap to restart

Controls

Space / ArrowUp / Tap jump · P / Esc pause · no double jump · Space restarts