Yaswanth A.

Key Skills

Software

AWS SageMaker

Appen

Anno-Mage

Axiom AI

Clickworker

CloudFactory

CrowdSource

CrowdFlower

Dataloop

Data Annotation Tech

Datatroniq

Datumbox

Diffgram

Doccano

Figure Eight

Hasty

Humanatic

iMerit

Label Studio

LabelImg

Mercor

Micro1

Mighty AI

OneForma

OpenCV AI Kit (OAK)

Sama

Sloth

Snorkel AI

Toloka

Telus

VoTT

Top Subject Matter

Internal research tooling for finance/analyst workflows (sentiment, macroeconomic, fundamental research)

Generative image modeling and synthetic dataset creation for controlled SDXL generation

Top Data Types

Document

Text

Image

Top Task Types

Prompt + Response Writing (SFT)

Fine-tuning

Classification

Bounding Box

Segmentation

Polygon

Entity (NER) Classification

Point/Key Point

Polyline

Cuboid

Object Detection

Text Generation

Question Answering

Text Summarization

RLHF

Red Teaming

Transcription

Evaluation/Rating

Computer Programming/Coding

Data Collection

Function Calling

Freelancer Overview

AI Engineer focused on RAG-based LLM research assistants (Piper Sandler). Brings 5+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include LangChain, Pinecone, and FAISS. Education includes Master of Science, University of Cincinnati. AI-training focus includes data types such as Document, Images, and Text and labeling workflows including Prompt + Response Writing (SFT), Fine-tuning, and Sentiment Analysis.

Labeling Experience

AI Engineer - Piper Sandler

TextClassification

Developed LLM-powered internal research tools using Retrieval-Augmented Generation (RAG) to support analyst workflows across sentiment, macroeconomic, and fundamental research. Improved LLM response quality through prompt engineering, document chunking strategies, and embedding-based retrieval for more relevant and consistent insights. Collaborated with ML and backend engineers to integrate LLM services via REST APIs and support scalable pipelines, model evaluation, and continuous performance improvements. • Built RAG workflows with LangChain and Pinecone/FAISS for contextual research and Q&A. • Designed agent-style task orchestration and context management for multi-step analytical processes. • Integrated LLM service endpoints and supported iterative model evaluation and tuning. • Implemented retrieval and prompt strategies to enhance relevance, consistency, and output quality.

2025 - 2025

AI Engineer focused on RAG-based LLM research assistants (Piper Sandler)

DocumentPrompt + Response Writing (SFT)

Built LLM-powered internal research tooling using Retrieval-Augmented Generation (RAG) to generate contextual insights from unstructured analyst materials. Implemented prompt engineering, document chunking strategies, and embedding-based retrieval to improve relevance and consistency of generated outputs. Collaborated with ML and backend engineers to integrate LLM services via REST APIs and support scalable pipelines for model evaluation and iterative improvements. • Used RAG with LangChain and vector indexing via Pinecone/FAISS. • Designed agent-style multi-step orchestration and context management for analytical workflows. • Worked on model evaluation processes tied to generated research responses. • Supported performance improvements through iterative integration of LLM services.

2025 - 2025

AI/ML Engineer building and fine-tuning SDXL Multi-LoRA pipelines with synthetic training data (SAP)

Fine-tuning

Designed and deployed a controllable SDXL image generation pipeline using Multi-LoRA and composable diffusion techniques to improve fidelity and regional control. Built and maintained a 25K-image synthetic training dataset using automated vision-language captioning and parallel data pipelines to enhance fine-tuning stability and data consistency. Fine-tuned custom SDXL LoRA models and performed systematic hyperparameter optimization to improve consistency and generalization across diverse prompts. • Created synthetic image-caption training data using LLaVA-13B and BLIP-2 captioning. • Applied Noise Offset and Min-SNR weighting to reduce training instability. • Tuned LoRA rank, learning rate, and mixing ratios for style/character consistency. • Integrated SAM with Stable Diffusion inpainting and evaluated ControlNet/IP-Adapters for localized editing coherence.

2022 - 2024

Education

Work History