Noveum.ai
Noveum is an AI-agent reliability and evaluation platform for production chat, voice and workflow agents. It connects instrumented tracing, configurable evaluation, synthetic testing, guardrails and evidence-linked recommendations so engineering teams can find failures, decide what to change and verify the result.
At a Glance
- Enterprise AI engineering and platform teams
- Companies operating production chatbots and customer-service agents
- Voice-agent and call-center platforms
- Financial-services, telecom and real-estate organizations
- +1 more
AI Tools by Noveum.ai
(1)
NovaSynth by Noveum
Synthetic User Testing for Voice Agents
Discussions
No discussions yet
Be the first to start a discussion about Noveum.ai
Latest News
NovaSynth launched on Product Hunt and ranked #5 Product of the Day.
Noveum published “Hallucination Detection in Production AI Agents: Methods That Actually Scale.”
Noveum introduced NovaGuard in beta for model-spend caps and request/token rate limits enforced by a connected data plane.
Noveum expanded its product loop with NovaPilot, an evidence-linked analysis and recommendation workflow for evaluated failures.
Products & Services
Open-source Python tracing SDK for LLM calls, tools, retrievals and agent steps as connected spans. It supports custom spans, threads and multi-agent graphs, token/cost/latency metadata, audio and image artifacts, fail-soft transport and OpenTelemetry-aligned concepts.
Evaluation product that applies selected compatible scorers to configured dataset items. It offers 112 published scorers across 15 categories, including hallucination, RAG, answer quality, tool use, safety and voice/audio metrics.
AI-assisted diagnosis and recommendation layer that analyzes evaluated datasets, groups recurring failure patterns, links findings to traces/items and ranks recommendations by priority, confidence, effort and expected impact.
Synthetic-user testing for voice and chat agents. It runs personas and scenarios over real SIP calls, LiveKit audio or chat, exercising interruptions, accents, adversarial intent and other edge cases, then captures and scores the runs.
Market Position
Noveum positions itself as a closed-loop AI-agent reliability platform rather than only an observability dashboard or an evaluation API: it connects production traces to configurable scorers, synthetic voice/chat testing and evidence-linked recommendations. Its comparison materials identify Langfuse, Arize/Arize Phoenix, Braintrust, AgentOps and Confident AI as relevant alternatives; the open-source SDK, real telephony testing and combined trace-to-recommendation workflow are differentiators.
Leadership
Founders
Shashank Agarwal
Founder and CEO. Previously a member of the original AWS SageMaker launch team and worked on personalization at Amazon Prime Video; earlier founded PipFeed and worked with Activeloop, Levity, Expedia and Hopdata.
Aditi Upadhyay
Co-founder and Chief Growth Officer. Previously a founding team member at Playo, helping grow it from zero to millions of users, and part of the founding team at Singapore fintech Spenmo; she describes her earlier work as AI-adoption strategy.
Executive Team
Shashank Agarwal
Founder & CEO
Former AWS SageMaker launch-team member and Amazon/Prime Video personalization engineer; previously founded PipFeed.
Aditi Upadhyay
Co-founder & Chief Growth Officer
Founding team member at Playo and Spenmo, with prior AI-adoption strategy experience.
Founding Story
Noveum says it began from the founders' frustration that an LLM agent can be built quickly for a demo but is difficult to keep reliable in production. Traditional APM and logs do not explain plausible-but-wrong agent behavior, so Noveum set out to build an AI-native reliability layer that connects traces, evaluation, synthetic testing and actionable, verifiable recommendations.
Business Model
Revenue Model
Organization-flat SaaS subscriptions based on monthly credits, trace/span capacity, attachment storage and included members; enterprise customers can purchase custom plans and scoped deployment support. NovaSynth testing, evaluation scoring and NovaPilot analysis draw from a unified credit system.
Pricing Tiers
2.5K credits/month, 1M spans/month, 2 GB attachment storage, 3 members.
10K credits/month, 20M spans/month, 20 GB storage, 10 members.
25K credits/month, 50M spans/month, 50 GB storage, 15 members.
50K credits/month, 100M spans/month, 100 GB storage, 25 members.
200K credits/month, 500M spans/month, 500 GB storage, unlimited members.
Unlimited credits, spans, storage and members; contact sales.
Target Markets
- Enterprise AI engineering and platform teams
- Companies operating production chatbots and customer-service agents
- Voice-agent and call-center platforms
- Financial-services, telecom and real-estate organizations
- Teams building RAG, LLM and multi-agent applications
- Production monitoring and debugging of chatbots and LLM applications
- Evaluation of RAG quality, hallucination, answer correctness, tool use and safety
- Pre-production voice-agent rehearsal and regression testing
- Testing multi-agent and workflow-agent routing, tools and outcomes
- Load and concurrency testing for voice agents
- Turning production failures into repeatable datasets and engineering recommendations
- MyOperator
- DarGlobal