Scientific writing, automated
LLM-powered report generation for research workflows — multi-model comparison, semantic search, and automated evaluation, in daily production use by scientists.
●Applied GenAI · San Francisco Bay Area
I build AI products that actually ship.
I'm a Data Scientist in Applied Generative AI at Bayer Crop Science R&D, where I design and deploy LLM-powered tools that scientists use every day. I also founded Phirky.ai — an AI day-planner for families, built solo end to end and live since July 2026.
Ph.D. in Computer Engineering from UMass Amherst; 17 peer-reviewed publications spanning computer vision, satellite intelligence, and trustworthy AI. One through-line across a decade: turning messy real-world signals into decisions people act on — satellite pixels, weather fields, crop R&D data, and now a family's Saturday.
Featured launch — founder work
Live · Launched July 2026
Phirky.ai
Planning a toddler weekend used to take twelve open tabs. Phirky turns one plain-English prompt into a ranked, routable day plan for Bay Area families — chaining activities by child age, weather, budget, proximity, and nap-time logistics.
Solo 0-to-1: product, engineering, brand, and launch. React/Vite on Vercel, FastAPI on Railway, Claude API + Google Places over a verified Bay Area place graph.
Plan a day ↗Rainy Saturday, toddler under 3, under $20, near Daly City
↓ becomes
9:30a — Indoor play café, stroller-friendly
11:15a — Toddler storytime, 6 min away
12:30p — Kid-approved lunch spot en route home
One prompt. Zero tabs.
At work — applied GenAI in production
Bayer Crop Science R&D
Report generation
Agent systems
Natural-language data access
Agent-based interfaces that let scientists query complex R&D datasets in plain English — designed for correctness and trust over flash.
Developer tooling
Codebase intelligence
LLM tooling that reads and explains unfamiliar codebases, cutting researcher onboarding onto legacy analytics projects from weeks to hours.
Conversational memory
Assistants that remember
Hierarchical short- and long-term memory for an organization-wide natural-language assistant — leading a small engineering team on design and delivery.
The log — recently shipped
-
LaunchedPhirky.ai is live — an AI day-planner for Bay Area families. One prompt in, a full ranked itinerary out.
-
ShippedThird major release of an LLM report-generation platform at work — now spanning multiple models with built-in evaluation.
-
SpokeInternal workshop talk on building, comparing, and shipping with LLM APIs — beyond the chat box.
Research — where the rigor comes from
My research arc has always pointed at the planet: sustainability first — solar forecasting and modeling from multispectral satellite imagery during my Ph.D. — then weather, extracting meteorological patterns and building trustworthy AI for the environmental sciences as a postdoc, and now food security, applying generative AI to crop science R&D. Along the way: diffusion-based foundation models for unlabeled satellite data and 17 peer-reviewed publications. Selected work below.
Trust and Trustworthy Artificial Intelligence: A Research Agenda for AI in the Environmental Sciences
Bostrom, Bansal, et al. — Risk Analysis (Wiley), 2023
Paper ↗
Leveraging Spatiotemporal Information in Meteorological Image Sequences: From Feature Engineering to Neural Networks
Environmental Data Science Journal (Cambridge), 2023
Paper ↗
A Moment in the Sun: Solar Nowcasting from Multispectral Satellite Data using Self-Supervised Learning
ACM e-Energy, 2022
Paper ↗
See the Light: Modeling Solar Performance using Multispectral Satellite Data
ACM BuildSys, 2020 — honorable mention, best presentation
Paper ↗
All publications
Estimating Low-Level Water Vapor from GOES ABI using Machine Learning
Lee, Hilburn, Bansal — journal, under review (2024)
Exploiting Satellite Data for Solar Performance Modeling
IEEE SmartGridComm, 2020
Paper ↗
On the Feasibility, Cost, and Carbon Emissions of Grid Defection
IEEE SmartGridComm, 2019 — Beijing, China
Paper ↗
SunShade: Enabling Software-Defined Solar-Powered Systems
ACM/IEEE ICCPS, 2017
Paper ↗
Self-Supervised Learning on Multispectral Satellite Data for Near-Term Solar Forecasting
ICML Workshop on Tackling Climate Change with ML, 2021
Paper ↗
Artificial Intelligence for Low-Level Moisture from GOES-R Series
103rd AMS Annual Meeting, 2023
Talk ↗
A Primer on Neural Network Architectures to Extract Information from Meteorological Image Sequences
103rd AMS Annual Meeting, 2023
Talk ↗
Distributed Health Monitoring System for Control in Smart Grid Network
IEEE ISGT-Asia, 2013
Paper ↗
Vulnerability Analysis of Power Grid Network against Failures by State Classification
IEEE EPEC, 2013
Paper ↗
Two-Tier Communication Architecture for Smart Meter
IEEE COMSNETS, 2013 — poster
Poster ↗
Health Monitoring in Smart Grid Using Big Data Perspective
IBM I-CARE, 2014
Ph.D. Thesis — Data-Driven Control, Modeling, and Forecasting for Residential Solar Power
UMass Amherst, 2021
Thesis ↗
Invited talks
Selected venues
Stanford University · Lawrence Berkeley National Laboratory · PARC (Xerox), Palo Alto · Oak Ridge National Laboratory GeoAI · UW–Madison Energy Analysis & Policy · ESST Transformers Workshop · AMS Annual Meeting.
Teaching & mentoring
Designed & taught
Created and independently delivered "Staring at the Sun: An Introduction to Solar Energy" (ENGIN 191) at UMass Amherst — and mentoring junior scientists and interns ever since.
Recognition
Awards
Honorable mention, best presentation — ACM BuildSys 2020 · 2nd prize, Duke Energy Data Analytics Lightning Talks 2020.
The journey — so far
- Jun 2023 — PresentData Scientist, Applied Generative AIBayer Crop Science R&D
- 2023Research ScientistGro Intelligence
- 2022 — 2023Machine Learning Scientist (Postdoc)CIRA, Colorado State University & NSF AI2ES
- Summer 2021Research Intern, GeoAI GroupOak Ridge National Laboratory
- 2015 — 2021Ph.D., Computer EngineeringUniversity of Massachusetts Amherst
- 2014 — 2015Research Associate, School of Computer ScienceNanyang Technological University, Singapore — real-time operating systems & smart batteries for vehicles
- 2011 — 2014M.S. by Research, Information TechnologyIIIT Bangalore