BMS College of Engineering

Bengaluru

Bachelor of Engineering

Computer Science

Aug 2019 – Jun 2023

Sarvam AI

Bengaluru

AI Engineer

Jan 2026 – Jun 2026

• Built pipelines and frameworks to automatically optimize agentic harnesses.
• Developed code-generation agents for custom coding workflows.
• Scaled scraper-authoring agents to generate scrapers on demand through the Model Context Protocol (MCP).

Sarvam AI

Bengaluru

AI Engineering Intern

Jul 2025 – Jan 2026

• Led an agentic scraper-generation system that reduced authoring time from three hours to 15 minutes, a 92% reduction across four clients.
• Architected subagent isolation to prevent context corruption and enable reliable code generation.
• Implemented automated API detection through runtime testing, then extended the framework to web automation and browser-operator agents.
• Deployed production workflows with Google Cloud Platform (GCP) Cloud Workflows and Cloud Scheduler.

Lossfunk

Bengaluru

AI Resident, Founding RecLLM

Apr 2025 – Jul 2025

• Founded RecLLM, an agentic recommendation system for conversational product discovery.
• Built adaptive memory to retain user preferences across long-running interactions.
• Shipped a production-ready Shopify prototype with automatic item indexing after validating business pain points.

Indian Institute of Science

Bengaluru

Research Assistant

Sep 2023 – Apr 2025

• Led an IBM Research collaboration on time-series imputation and developed a novel MLP Mixer architecture.
• Improved on state-of-the-art results by 21% with a 12K-parameter design and constant rather than quadratic parameter growth.
• Co-developed a self-diagnosing bias-mitigation framework using Neural Tangent Kernel attribution, vision-language model descriptions, and CLIP embeddings, improving worst-group accuracy by 10.6%.

Indian Institute of Science

Bengaluru

Research Intern

Jan 2023 – Sep 2023

• Co-designed a weak-supervision architecture combining a generative adversarial network, classifier, and label model, improving label accuracy by 3% and Fréchet Inception Distance by seven points.
• Built an OpenReview literature-survey tool and an automated experiment-management framework.

CryptoRelief

Bengaluru

Python Developer

May 2021 – Jul 2021

• Built APIs and SHA-256 search for COVID-19 resource tracking, delivering 90% faster retrieval for intensive-care, oxygen, and ambulance data.
• Created web scrapers and a Twitter bot for rapid information dissemination.

ELITS: Efficient Lightweight Imputation for Time Series

2026

Pranav Sastry, Kalyan Reddy, Sumanta Mukherjee, Vijay Ekambaram, Pankaj Dayama, and Prathosh AP

AAAI 2026 Workshop on AI for Time Series Analysis

Feature-Guided Subset Selection: Mitigating the Impact of Spurious Features via Data Attribution

2025

Kumar Shubham, Pranav Sastry, and Prathosh AP

NeurIPS 2025 Workshop on Reliable ML from Unreliable Data

Fusing Conditional Submodular GAN and Programmatic Weak Supervision

2024

Kumar Shubham, Pranav Sastry, and Prathosh AP

AAAI Main Technical Track 2024

Performance Analysis of Object Detection Algorithms for Waste Segregation

2023

Naman Singh, Pranav Sastry, Niharika B. S., Arka Sinha, and Umadevi V.

IEEE ICAISE 2023

Agentic Scraper Authoring System

Sarvam AI

• Generated and runtime-validated production scrapers from natural-language requirements, using network monitoring to discover backend APIs for bulk execution.

Time-Series Imputation

IBM Research

• Designed parallel MLP Mixers that improved on state of the art by 21% with 858× fewer parameters.

Weak Supervision

Indian Institute of Science

• Unified a generative adversarial network, classifier, and label model, improving label accuracy by 3% and Fréchet Inception Distance by seven points; published at AAAI 2024.

Data Attribution for Spurious Correlations

Indian Institute of Science

• Combined Neural Tangent Kernel attribution, vision-language descriptions, and CLIP embeddings to improve worst-group accuracy by 10.6%.

Object Detection for Waste Segregation

Engineering Major Project

• Co-led YOLO model design and built ODDGen to synthesize hundreds of thousands of annotated images; published at IEEE ICAISE 2023.

Aiter: Terminal AI Chat with Multi-Agent Support

• Built modular multi-agent chat with MCP integration, persistent sessions, and extensible agent scaffolding.

Neograd: Deep Learning Framework from Scratch

• Built and published a pure-Python automatic-differentiation framework with broadcasting, custom operations, and gradient checking.

Web Development and APIs

• Built an OpenReview paper scraper (https://github.com/pranftw/openreview_scraper), React post editor, bilingual Flask website (https://gururajhr.com), Twitter API bot (https://github.com/pranftw/twttr-bot), and GitIt (https://github.com/pranftw/gitit) for GitHub automation.