Custom > Agentic AI & ML Solutions
We automate manual processes with AI agents that understand your business logic, collaborate with your team, and learn from your expertise. We draw on 8+ years of AI/ML engineering and 20+ years of architecting, developing, and operating production software systems at scale[1].
Solutions we've built for clients serve thousands of B2B customers worldwide. We are active open source contributors[2], currently focused on multi-party conversational AI and on agents that act and self-improve via code actions._
automation
Grounded in Your Business Logic
Agents grounded in company-specific process knowledge and decision logic.
We extract and structure process knowledge from your existing sources (documentation, messaging histories, internal tools, expert interviews) and encode it into agent-accessible decision logic. Agents ground their outputs in this company-specific context, with every decision traceable to source material rather than inferred from general-purpose training data.
Grounded in Your Business Logic
We extract and structure process knowledge from your existing sources (documentation, messaging histories, internal tools, expert interviews) and encode it into agent-accessible decision logic. Agents ground their outputs in this company-specific context, with every decision traceable to source material rather than inferred from general-purpose training data.
Embedded in Your Workflows
Deployed into existing tools with graduated autonomy and escalation logic.
We deploy agents directly into your existing tools (Slack, email, ticketing systems, approval chains) rather than introducing separate interfaces. Agents follow a graduated autonomy model: starting in supervised mode where every action is reviewed, then progressing to autonomous handling of routine decisions with rule-based escalation when confidence is low or stakes are high.
Embedded in Your Workflows
We deploy agents directly into your existing tools (Slack, email, ticketing systems, approval chains) rather than introducing separate interfaces. Agents follow a graduated autonomy model: starting in supervised mode where every action is reviewed, then progressing to autonomous handling of routine decisions with rule-based escalation when confidence is low or stakes are high.
Learning from Your Feedback
Structured feedback loops that generalize feedback to future decisions.
Agents incorporate your team's feedback through structured feedback loops: approved, rejected, or edited outputs become training signals that generalize to similar future decisions. This keeps agent behavior aligned as business rules change, edge cases accumulate, and team knowledge evolves, without requiring manual reprogramming for each new scenario.
Learning from Your Feedback
Agents incorporate your team's feedback through structured feedback loops: approved, rejected, or edited outputs become training signals that generalize to similar future decisions. This keeps agent behavior aligned as business rules change, edge cases accumulate, and team knowledge evolves, without requiring manual reprogramming for each new scenario.
open_source
Agent harness and CLI tool that acts and self-improves via code actions.
Lightweight agent harness and CLI tool that acts by executing code and shell commands in a local, sandboxed IPython kernel. Uses code actions to improve itself and its tool library from experience. Calls MCP server tools programmatically, enabling composition of tools in a single inference pass.
Lightweight agent harness and CLI tool that acts by executing code and shell commands in a local, sandboxed IPython kernel. Uses code actions to improve itself and its tool library from experience. Calls MCP server tools programmatically, enabling composition of tools in a single inference pass.
Python code execution sandbox with programmatic MCP tool calling support.
Python code execution sandbox with first-class support for programmatic MCP tool calling. Generates typed Python APIs from MCP server schemas and executes code in a stateful IPython kernel. Features programmatic tool call approval workflows and lightweight sandboxing via Anthropic's sandbox-runtime.
Python code execution sandbox with first-class support for programmatic MCP tool calling. Generates typed Python APIs from MCP server schemas and executes code in a stateful IPython kernel. Features programmatic tool call approval workflows and lightweight sandboxing via Anthropic's sandbox-runtime.
Projects that integrate single-user AI agents into multi-party conversations.
Projects for integrating single-user AI agents into group conversations without requiring agent modification. Group Reasoners detect relevant patterns in multi-party message streams and transform them into self-contained queries for downstream AI agents. Agents act on behalf of individual group members, with connectors available for Slack and GitHub.
Projects for integrating single-user AI agents into group conversations without requiring agent modification. Group Reasoners detect relevant patterns in multi-party message streams and transform them into self-contained queries for downstream AI agents. Agents act on behalf of individual group members, with connectors available for Slack and GitHub.
clients
AI-powered visual search scaling to thousands of B2B customers.
Development of an AI-powered visual and hybrid search platform built as a cloud-native solution scaling to 2000+ B2B customers worldwide. Training of custom query processing models with synthetic data to deliver accurate search results across diverse content types.
Development of an AI-powered visual and hybrid search platform built as a cloud-native solution scaling to 2000+ B2B customers worldwide. Training of custom query processing models with synthetic data to deliver accurate search results across diverse content types.
Multimodal, agentic AI search engine powering the AP Newsroom.
Conception and development of a multimodal, agentic AI search engine with semantic understanding of images, videos, documents, and audio, enhanced through domain-specific model fine-tuning for media organizations. Powers the Associated Press Newsroom, and contributed to MerlinOne's successful acquisition by Canto in 2023.
Conception and development of a multimodal, agentic AI search engine with semantic understanding of images, videos, documents, and audio, enhanced through domain-specific model fine-tuning for media organizations. Powers the Associated Press Newsroom, and contributed to MerlinOne's successful acquisition by Canto in 2023.
Global DAM distribution and content playout for Red Bull TV streaming.
Global distribution of the in-house DAM system to support low-latency local access to content and write-availability under network partitions. Release of the underlying algorithms and protocols as open source project. Content playout platform for Red Bull TV's global streaming service, delivering content to millions of users worldwide with consistent performance.
Global distribution of the in-house DAM system to support low-latency local access to content and write-availability under network partitions. Release of the underlying algorithms and protocols as open source project. Content playout platform for Red Bull TV's global streaming service, delivering content to millions of users worldwide with consistent performance.