Tanishka A.

Key Skills

Software

Internal/Proprietary Tooling

Top Subject Matter

LLM alignment

Safety Domain Expertise

and supervised fine-tuning dataset preparation

Top Data Types

Text

Audio

Video

Top Task Types

RLHF

Transcription

Classification

Text Generation

Question Answering

Text Summarization

Red Teaming

Prompt + Response Writing (SFT)

Freelancer Overview

Prompt Engineer – Soul AI (Model Alignment, CoT & SFT). Brings 3+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Internal, Proprietary Tooling, and Google Cloud Vertex AI. Education includes Bachelor of Technology, Gautam Buddha University (2024). AI-training focus includes data types such as Text and Audio and labeling workflows including RLHF and Transcription.

Labeling Experience

Prompt Engineer – Soul AI (Model Alignment, CoT & SFT)

TextRLHF

Performed reinforcement learning from human feedback (RLHF) to rank model outputs for technical accuracy, logical coherence, and safety, improving alignment with human intent. Engineered supervised fine-tuning (SFT) training datasets using LaTeX-audited chain-of-thought solutions to support multi-step reasoning. Conducted evaluation and quality strengthening through safety-centric workflows. • Ranked generated outputs using human-feedback criteria (accuracy, coherence, safety). • Built SFT datasets with LaTeX-audited chain-of-thought reasoning. • Ran red-teaming protocols to surface edge-case vulnerabilities and hallucinations. • Managed high-velocity training task execution with review/acceptance loops (1,000+ tasks).

2024 - Present

AI/ ML Founding Intern – Arial Labs

Google Cloud Vertex AIAudioTranscription

Built a low-latency streaming speech-to-text pipeline to support accurate downstream language-model processing. Engineered a system to isolate dual-channel audio streams for more reliable transcription in live meeting contexts. Added filtering and automated behavioral scoring against human-reviewed transcripts to improve transcript quality and response suitability. • Created WebRTC/WebSocket infrastructure to separate dual-channel audio in Google Meet sessions. • Implemented asynchronous chunking transcription to reduce processing time and add caption/token filtering. • Designed a real-time classifier to score responses into qualitative tiers validated on 50+ manually reviewed transcripts. • Integrated cloud services for transcription and LLM-based processing to enable iterative improvements.

2024 - 2024

Education

Work History