Esha Joshi

Created DSPy prompt optimization workflows using BootstrapFewShot, COPRO, and Chain-of-Thought for NLP tasks. Improved resume extraction accuracy from 56.8% to 83.2% (p < 0.0001) and medical NER from 80.0% to 80.6%. Trained LightGBM forecasting models for different KPIs on 10M+ ad, adset and campaign records with 289 features, achieving test R2 > 0.90.

Software Engineering Intern at ThirtySignals (AIR OAK THIRTY LLP) (2026-02 – Present)

• Created DSPy prompt optimization workflows using BootstrapFewShot, COPRO, and Chain-of-Thought for NLP tasks.
• Improved resume extraction accuracy from 56.8% to 83.2% (p < 0.0001) and medical NER from 80.0% to 80.6%.
• Trained LightGBM forecasting models for different KPIs on 10M+ ad, adset and campaign records with 289 features, achieving test R2 > 0.90.
• Productionized Dagster 6h/24h forecasting pipelines, reducing processing time by 60% via batch processing.
• Added an LLM decision layer to convert predicted KPIs into proper recommendations for ad, adset and campaign.
• Designed TypeScript domain models and Zod-based YAML validation for the Pupil evaluation framework, catching config errors, duplicate IDs, and invalid scenarios before they reached IRIS agent evaluation runs.

Data Science & Data Analytics Intern at Pristine Forests (2025-11 – 2026-03)

• Created automated ETL pipelines using Airbyte, Python, and SQL, reducing manual processing time by 30% through custom field creation.
• Integrated Meta and Instagram Graph APIs to extract 8+ KPIs and built dashboards monitoring 10,000+ daily data points.
• Deployed Dockerized data pipelines and Apache Superset dashboards supporting 20+ KPI metrics across 5+ campaigns.

AI & Machine Learning Intern at Infosys Springboard (2025-01 – 2025-03)

• Implemented a real-time object recognition system using YOLO-based deep learning models.
• Trained and evaluated models on a custom dataset achieving 92% detection accuracy across 80 object categories.
• Deployed a computer vision pipeline using Python and OpenCV for real-time inference and live camera integration on streamlit.

B.Tech in Electronics and Communication Engineering – Indian Institute of Information Technology (IIIT) Pune (2022-11 – 2026-05)