Open frontier
Time-series
No dominant foundation model yet. Teams still rebuild from scratch for every domain, cadence, and signal type.
Enterprise time-series platform
Unified time-series intelligence
across enterprise data.
Samay AI is a platform — not a point model — for adapting foundational time-series intelligence to new domains with far less data, far less engineering, and far less infrastructure than bespoke pipelines or vertical-only tools.
Up to 0%
More data-efficient on new domains
0%
More accurate than domain-specific models
Over 0×
Smaller than competitors
Our motivation
Foundation models won language and vision.
Time-series is still wide open.
Every industry runs on temporal signals — demand, risk, energy, health — yet most organizations still maintain bespoke forecasting pipelines per use case. We believe the next wave of AI depends on a shared foundation for time-series intelligence.
Solved
Language & NLP
Foundation models already generalize across text tasks with minimal per-application engineering.
Solved
Computer vision
Transfer learning works across image domains — a proven template for what time-series still lacks.
Why time-series is uniquely hard
No fixed semantics
The same metric means different things at different cadences and contexts — daily vs. weekly is not just resampling.
Regime changes
Distributions shift mid-stream. Models must adapt to structural breaks, not fail silently when the world changes.
Pattern diversity
One model must span trend, seasonality, anomalies, and volatility — often within a single series.
Our vision
One platform for unified time-series intelligence — forecasting, anomaly detection, classification, and decision support across finance, supply chain, healthcare, energy, retail, and IoT. One adaptable layer instead of thousands of brittle pipelines.
One adaptable platform for time-series intelligence across every industry — instead of thousands of brittle, single-use pipelines.
The platform
Data-efficient intelligence that generalizes — not another vertical model.
Samay is the platform layer for enterprise time-series: ingest messy signals, adapt a shared foundation to new domains with minimal data, and deploy behind your security perimeter.
The platform layer for enterprise time-series — adapt a shared foundation to any domain and deploy inside your perimeter.
01
Data-efficient domain adaptation
Reach production quality on new domains with up to 60% less data than training from scratch — built for enterprises that can't wait years before every use case pays off.
02
General beats domain-specific
One shared foundation outperforms vertical-only models — without a separate pipeline per team, SKU, or business unit.
03
Compact, deployable footprint
The engine behind the platform is over 2× smaller than competing approaches — practical for batch inference inside VPCs and on-prem.
04
Adaptive tokenization
Learns time-series vocabulary as semantic tokens and adapts to regime changes — core research from Kamarthi & Prakash at NeurIPS 2024.
05
Weeks to minutes
From historical enterprise data to production-ready intelligence in a fraction of the time — not months of bespoke pipeline engineering.
06
Enterprise-grade deployment
Built for environments where data cannot leave the perimeter — private deployment, predictable inference, guardrails for regulated industries.
Founding team
Decades at the edge of time-series AI.
A world-class founding team spanning foundational ML research, large-scale industry deployments at the highest level of scrutiny.
Two founders with 20+ years combined in foundational ML research and time-series AI, with experience at leading institutions (Georgia Tech, CMU, IIT). Published at leading venues (NeurIPS, ICML, ICLR, Nature family, PNAS), with $50M+ in collaborations across big tech, the Fortune 500 (Meta, Google, Walmart, Samsung Research, MS Research, Dow, Conviva, CDC, NIH).
Harshavardhan Kamarthi
Co-founder
- Final-year ML PhD, Georgia Tech; IIT Madras (Computer Science)
- Lead inventor of the Samay foundation model engine
- Published at NeurIPS, KDD, ICLR, WWW, and Nature journals
- Industry experience at Samsung, Conviva, and Dow
- Led a top-ranked team in the CDC forecasting initiative — building state-of-the-art models in a high-stakes public-health competition referenced by media and the CDC director
Dr. B. Aditya Prakash
Co-founder
- Full Professor & Associate Chair, Georgia Tech; PhD from CMU, B.Tech from IIT Bombay
- $50M+ in research collaborations across Meta, Walmart, Google, CDC, NIH, and NSF
- NSF CAREER, IEEE AI's 10 to Watch, NAE Frontiers of Engineering
- Lead author on adaptive time-series tokenization (NeurIPS 2024)
- Decades of experience in scalable AI, time-series, and graph learning systems
Past research & industry collaborations
These organizations reflect prior academic and industry work — not customers, partners, or endorsements of Samay AI.
Get in touch
Let's explore what temporal intelligence can do for you.
Whether you're exploring an enterprise pilot, a research collaboration, or an investment conversation — we'd like to hear from you.