Agentic AI • ML Systems • Quantitative Modeling
Machine learning for trading desks.|
I build ML systems that survive contact with production: demand forecasting on a proprietary trading desk, Boeing funded research on agentic AI for engineering design, and pipelines that run end to end from raw data to deployment. The domains vary; the standard does not. Evidence over vibes, tests over demos.
Focus
LLM agents, RAG, evaluation, data engineering, cloud deployment
Stack
Python, Spark, GCP, PostgreSQL, Docker, Next.js, Mapbox
Now
Quant work in power markets + Boeing funded research on MCP agent tooling
> location: Pittsburgh, PA
> desk time: ··:··:·· ET
> status:open to full time roles, Dec 2026
> current: CMU + Boeing funded MCP research
> grid frequency: 60.00 Hz
> last shipped: 8d ago, mayank-portfolio
> uptime: 3 publications, 0 unhandled exceptions▊
Experience
My work through the years
Most recent first. Parallel bars mean parallel lives.
Summer 2026
Quantitative Analyst Intern, Load Forecasting
Five Dimensions Energy · Princeton, NJ
Four solo projects on a PJM trading desk: production day ahead forecasts, a real time zonal forecast, data center load intelligence.
2025 → present
M.S. Artificial Intelligence Engineering
Carnegie Mellon University · GPA 3.92, Pittsburgh, PA
Energy Science, Technology and Policy track. Graduating Dec 2026.
Graduate Research Assistant
CMU Design Research Collective · Boeing funded
MCP agent pipelines that run aircraft design physics end to end. Open source under cmudrc.
2023 → 2025
Data Scientist → Senior Data Scientist
Gentari (PETRONAS) · Gurgaon, India
Production solar forecasting that beat vendor accuracy, a private RAG assistant, automation across 40 renewable plants.
2022 → 2023
Research Assistant, B.Tech thesis
NSUT Delhi
Bearing fault detection with Vision Transformers. Two peer reviewed papers.
Research Intern
IIT Kharagpur
Adversarial robustness for autonomous driving. Published at CVIP 2022.
2019 → 2023
B.Tech, Electrical Engineering
NSUT Delhi · IAFBA national scholar
Minor in ML and Data Science. First place of 110 teams at the IIT Ropar Hackathon along the way.
Same story, as a chart
The career graph
x axis: years. y axis: ambition. Hover the peaks or pick a year.
Quantitative Analyst Intern
Five Dimensions Energy2026Load forecasting on a PJM trading desk. Improved the production day ahead model, shipped a real time forecast for 20 zones, and mapped data center load growth.
A few work examples
Projects with real systems, data, and constraints
About
How I build
I like projects where the hard part is the system: messy data, real-time constraints, ambiguous objectives, and shipping something that holds up in production.
- • Build pipelines that are reproducible (tests, schemas, deterministic outputs)
- • Optimize for reliability first, then performance (profiling, caching, batching)
- • Measure quality with evals, not vibes (benchmarks, offline + online metrics)
INSERT COIN
Interactive
The arcade
Three games. Tech wordle: six tries at a five letter word from code and AI, with hints when you struggle. Bit flip: match the binary before the clock runs out. Beat the forecast: five days against the model, the game I played all summer minus the money.
Beyond work
Things I do when I’m not coding
I like building prototypes, traveling for conferences, getting out on the water, racing triathlons, and pointing snowboards down hills.
All the gear, some idea
Swim, bike, run. First triathlon in the books
Moon tailed grouper
King mackerel
Giant trevally
Through the lens
The soundtrack
On repeat, last 4 weeks
Radiohead
59 plays
Tame Impala
27 plays
Imagine Dragons
33 plays
The Strokes
19 plays
Coldplay
22 plays
Steve Lacy
22 plays
All time heavy rotation
Kanye West
4,820 plays
Tame Impala
2,432 plays
The Weeknd
2,609 plays
Kendrick Lamar
1,985 plays
Frank Ocean
1,413 plays
Radiohead
1,519 plays
Pulled from my Spotify history via stats.fm, refreshed weekly.