Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

I am Sonal Pardesi, an AI professional focused on Trust & Safety, policy operations, and LLM evaluation, currently supporting Meta's Generative AI team. I excel at handling high-risk and borderline content, driving policy-consistent decisions, and improving model reliability at scale through annotation, escalation analysis, and prompt testing. I am a collaborative problem solver who aligns policy, QA, and technical teams to deliver accurate, safe, and impactful results. I also provide market-specific policy insights and KPI-driven evaluation to inform governance and product decisions across multilingual content.…I am Sonal Pardesi, an AI professional focused on Trust & Safety, policy operations, and LLM evaluation, currently supporting Meta's Generative AI team. I excel at handling high-risk and borderline content, driving policy-consistent decisions, and improving model reliability at scale through annotation, escalation analysis, and prompt testing. I am a collaborative problem solver who aligns policy, QA, and technical teams to deliver accurate, safe, and impactful results. I also provide market-specific policy insights and KPI-driven evaluation to inform governance and product decisions across multilingual content.

Sonal Pardesi

AI Engineer, Data Scientist, Data Analyst, +2





I am Sonal Pardesi, an AI professional focused on Trust & Safety, policy operations, and LLM evaluation, currently supporting Meta's Generative AI team. I excel at handling high-risk and borderline content, driving policy-consistent decisions, and improving model reliability at scale through annotation, escalation analysis, and prompt testing. I am a collaborative problem solver who aligns policy, QA, and technical teams to deliver accurate, safe, and impactful results. I also provide market-specific policy insights and KPI-driven evaluation to inform governance and product decisions across multilingual content.…I am Sonal Pardesi, an AI professional focused on Trust & Safety, policy operations, and LLM evaluation, currently supporting Meta's Generative AI team. I excel at handling high-risk and borderline content, driving policy-consistent decisions, and improving model reliability at scale through annotation, escalation analysis, and prompt testing. I am a collaborative problem solver who aligns policy, QA, and technical teams to deliver accurate, safe, and impactful results. I also provide market-specific policy insights and KPI-driven evaluation to inform governance and product decisions across multilingual content.

Available to hire

I am Sonal Pardesi, an AI professional focused on Trust & Safety, policy operations, and LLM evaluation, currently supporting Meta’s Generative AI team. I excel at handling high-risk and borderline content, driving policy-consistent decisions, and improving model reliability at scale through annotation, escalation analysis, and prompt testing.

I am a collaborative problem solver who aligns policy, QA, and technical teams to deliver accurate, safe, and impactful results. I also provide market-specific policy insights and KPI-driven evaluation to inform governance and product decisions across multilingual content.

Skills

Experience Level

Expert

Expert

Expert

Expert

Expert

Intermediate

Intermediate

Intermediate

Intermediate

Intermediate

Language

English

Fluent

Hindi

Fluent

Marathi (Marāṭhī)

Advanced

Work Experience

Quality Measurement Specialist - LLM at Covalen OnSite / Meta Dublin

February 1, 2025 - Present

Annotated, reviewed, and evaluated multilingual LLM data (English, Hindi, Marathi) across 13 Trust & Safety categories; performed escalation and safety incident investigations on borderline content; designed adversarial and multi-turn prompts to stress-test model behavior; identified edge cases and contributed to guideline refinements to support scalable policy enforcement; shared structured quality feedback with vendor and internal teams; tracked remediation effectiveness and measured accuracy, consistency, and severity KPIs; conducted qualitative and quantitative error analysis to identify systemic gaps and calibrated reviewer decisions with policy stakeholders; delivered India-market insights on content trends and policy edge cases.

AI Trainer - Freelancer at Outlier AI

September 1, 2024 - January 31, 2025

Provided expert-level evaluations to improve LLM prompt understanding, bias detection, and policy-aligned outcomes; contributed to automation strategies for scalable human-in-the-loop moderation; designed and evaluated Python- and JavaScript-based LLM prompt workflows; tested prompt structures against real-world scenarios to refine model behavior and safety compliance.

Owner at Food Corp

December 1, 2020 - June 1, 2023

Led marketing, content, and data-driven growth initiatives, increasing customer base by 50% and revenue by 75%; analyzed customer feedback to improve retention and ratings, exceeding industry benchmarks.

Business Development Executive at Ashtech Infotech Pvt Ltd

January 1, 2018 - February 1, 2019

Engaged with enterprise clients and OEMs (AWS, Microsoft, Dell EMC) to deliver IT and cloud solutions with regulatory compliance; used Salesforce and JIRA for stakeholder reporting and issue resolution; maintained post-sales communication to flag policy exceptions and escalate unresolved compliance issues.

Business Development Executive at Handcraft Worldwide

May 1, 2017 - December 1, 2017

Managed website and online engagement to boost client navigation and inquiries; analyzed market trends, coordinated product updates, and handled proposals, invoices, and billing; led product development initiatives based on customer feedback, achieving sales growth.

Network Support Engineer at Impact Infotech Pvt Ltd

September 1, 2015 - May 1, 2016

Provided L1 IT support for Vodafone leased lines, maintaining 80% network stability; handled incidents via ticketing systems and ITIL processes; collaborated with internal teams to resolve technical challenges using Cisco technologies.