Ashr

Roadmap & Position in AI/ML Testing

Catches AI agent failures before users see them by stress-testing across text, voice, and images.

Company Overview

Builds an automated testing platform that mimics users in production environments to catch AI agent failures before they reach real users. Uses synthetic test generation and custom ML scorers to evaluate agents across multiple modalities.

What They're Building

The company's public product roadmap & what they're committed to building.

Testing platform mimicking users in production environments. Synthetic scenario-driven testing. Multi-modal testing across text, voice, and images.

Latest Intelligence

Zeitgeist tracks private signals to determine where the company is heading strategically.

Agent Testing Moves Into Production

May 13, 2026

Confidence:

High

New Intel: Ashr is stress-testing AI agents by mimicking users in production across text, voice, and images. That targets the reliability gap between impressive demos and agents trusted with real customers.

Founder and Key Execs

Shreyas Kaps, Co-Founder

Co-Founder

Rohan Kulkarni, Co-Founder

Co-Founder

Founder Force Multiplier

Shreyas Kaps brings 2 years building AI agents in production (finance and DevOps). Rohan Kulkarni (Berkeley EECS) has a successful exit (Ask Geri, acquired).

Funding History

2025 | Shreyas Kaps and Rohan Kulkarni co-found Ashr.
2026 | Accepted into Y Combinator W26 batch.

Competitors

AI Agent Testing

Patronus AI, Galileo AI, Confident AI (DeepEval).

General ML Testing

Weights and Biases, Arize AI, Kolena.

Emerging

AgentOps, LangSmith, Braintrust.

Ashr

's Moat:

Custom ML scorers trained on each customer's specific agent failure modes create switching costs. The synthetic test library grows with each deployment, building a regression suite that would take months to recreate on another platform.

How They're Leveraging AI

LLM synthetic scenario generation

Uses large language models to automatically generate diverse, realistic, and adversarial synthetic test scenarios that simulate real-world user interactions with AI agents across multiple modalities.

Custom ML evaluation scoring

Deploys fine-tuned machine learning models as custom scorers that automatically evaluate AI agent outputs against business-specific quality criteria, replacing subjective human review with consistent, scalable, and configurable evaluation.

Multi-modal swarm stress testing

Orchestrates large-scale, ML-driven swarm tests that simultaneously bombard AI agents with thousands of diverse multi-modal inputs to uncover rare edge cases, performance bottlenecks, and cross-modal failure modes that traditional testing misses.

AI Use Overview:

Using LLM-driven synthetic test generation, custom ML scorers for business-specific quality evaluation, and multi-modal swarm testing that uncovers rare edge cases at scale.

More Similar Companies

Arena (formerly LLMArena)

Crowdsourced human-preference benchmarking platform for LLMs and generative AI models.

Neutral third-party evaluation becomes critical infrastructure as model proliferation outpaces any single lab's ability to grade itself credibly.

Cajal

Deploys AI mathematicians that formally verify proofs, grounding outputs in truth not guesses.

LLMs hallucinate. Lean proves things. Cajal pairs LLMs with formal verification so every mathematical result is machine-checked, starting with quantum computing and finance where a wrong proof costs real money.

Cascade

Evaluates and certifies AI agents for safe deployment with red teaming and formal guarantees.

Red teaming and guardrails exist as separate tools. Cascade combines them into one platform with adaptive scaffolding that learns from production runs, already deployed across legal reasoning and customer support agents. The CEO researched graph reasoning and agentic safety at UC Berkeley's BAIR Lab.

Envariant

Lets model builders inspect and steer AI behavior inside the latent space to catch failures.

Most AI safety tools work on model outputs. Envariant operates inside the latent space itself, detecting hallucinations and drift at the representation level before they surface. Beta SDK launched with applications in text LLMs, robotic agents, and protein models.

Back To All Companies >