Satwik Satwik

Key Skills

Software

Scale AI

Labelbox

Top Subject Matter

No subject matter listed

Top Data Types

Text

Top Task Types

Evaluation Rating

Fine Tuning

Prompt Response Writing SFT

Question Answering

RLHF

Freelancer Overview

I bring a unique combination of hands-on product management and deep user insight to AI training and data labeling. Over the past 3 years, I’ve launched and grown B2B SaaS products and consumer-facing platforms from scratch, managing cross-functional teams and implementing structured feedback loops to refine product experiences. At Humantic AI, I was the first product manager and played a critical role in migrating backend systems from GPT-3 to GPT-3.5, which involved designing and managing prompt frameworks—a direct overlap with RLHF workflows. My work reduced email hallucinations by 80% and led to one of the product’s highest-rated features by NPS. In every role—whether scaling lead qualification systems at LeapFinance, deploying behavioral segmentation at BetterPlace, or rapidly iterating onboarding funnels at Raaz—I’ve relied on meticulous data labeling, user modeling, and feedback loops to improve precision and personalization. What sets me apart is my ability to fuse structured thinking with empathetic user insight, a skill crucial for training human-aligned AI systems. I’ve also led sales and growth teams, giving me hands-on experience in defining nuanced evaluation criteria—skills that translate directly into high-quality, context-sensitive RLHF contributions.

Labeling Experience

Email evaluation

LabelboxTextRLHFFine Tuning

Led a high-precision prompt evaluation and refinement initiative during the backend migration from GPT-3 to GPT-3.5 for a B2B SaaS email personalization platform. Using Labelbox, I systematically reviewed AI-generated emails for tone accuracy, factual alignment, personalization depth, and coherence. Each sample was annotated with structured qualitative feedback and ranked relative to others to train preference models (RLHF). I also iterated prompt variations and contributed human-written completions for supervised fine-tuning (SFT), with a focus on professional tone and context-sensitive personalization. The dataset covered 1,000+ email samples across diverse sales personas, industries, and use cases. Quality benchmarks were defined by internal bug triage metrics and end-user NPS. My contributions led to an ~80% reduction in buggy outputs and significantly improved downstream model quality as evidenced by customer feedback (5-star G2 reviews). The project adhered to internal QA loops

Led a high-precision prompt evaluation and refinement initiative during the backend migration from GPT-3 to GPT-3.5 for a B2B SaaS email personalization platform. Using Labelbox, I systematically reviewed AI-generated emails for tone accuracy, factual alignment, personalization depth, and coherence. Each sample was annotated with structured qualitative feedback and ranked relative to others to train preference models (RLHF). I also iterated prompt variations and contributed human-written completions for supervised fine-tuning (SFT), with a focus on professional tone and context-sensitive personalization. The dataset covered 1,000+ email samples across diverse sales personas, industries, and use cases. Quality benchmarks were defined by internal bug triage metrics and end-user NPS. My contributions led to an ~80% reduction in buggy outputs and significantly improved downstream model quality as evidenced by customer feedback (5-star G2 reviews). The project adhered to internal QA loops

2023 - 2023

Education

Work History