ArcellAI Memo

by the arcellai founding team — learn more at arcellai.com or view pitch deck

Founder Memo

Overview

ArcellAI is an operations research and applied AI engineering lab developing an interactive AI data scientist and engineer for mission-critical systems.

AI deployments fail or remain trapped in isolated sandbox environments because real-world systems are fragmented, disconnected, and lack a shared situational context layer. Defense, logistics, and mission-critical engineering operators waste valuable time performing manual data retrieval, custom scripting, and fragile spreadsheet assembly to answer operational questions.

The bottleneck is not model reasoning capacity; it is the lack of a secure context-engineering and provenance layer. Operators need trusted, traceable answers that link every recommendation, action, and reasoning step back to verified source telemetry, technical manuals, and historical logs.

Interactive context-aware agents change the equation. Suddenly, navigating disconnected databases, analyzing real-time sensor streams, and tracing dependencies across complex mission systems becomes an interactive natural language conversation with an agent that understands systems-engineering context, maintains precise provenance graphs, and executes authorized actions safely.

We're building the infrastructure where mission-critical systems engineering and defense operations move at software speed.

Real Example: Mission Logistics and Supply Chain Dependency Analysis

Instead of abstract promises, here's a concrete workflow run end-to-end using our interactive agent.

  • Objective: "Evaluate mission-readiness and supplier dependency risks across a distributed maintenance database and technical component manuals."
  • Data Sources: The agent autonomously ingests data from supplier tables (SQL/Sovereign data lakes), maintenance operational logs (timeseries streams), and technical manuals (unstructured PDF documents).
  • Execution:
    • Planner: Decomposes the request into logical steps: schema harmonization → technical document semantic search → dependency path tracing → supplier risk calculation.
    • Executor: Writes and executes precise code steps within a secure sandbox environment, resolving data discrepancies and handling errors.
    • Context Harness: Tracks the complete lineage of every relationship. It resolves that "Part_ID" from a local maintenance database matches "NSN_Code" in a government supplier ledger.
  • Result: A fully traceable dependency map, versioned in our Provenance Graph, with explicit links to source documents. What previously took days of manual coordination is resolved in minutes with absolute traceability.

Rigorous Validation and Academic Pedigree

PyTDC (Therapeutics Data Commons) serves as one of our key validation benchmarks for complex, high-stakes modeling. By integrating PyTDC's rigorous scientific standards with our agentic orchestration, we have validated our capability to handle highly specialized, high-dimensional datasets. ArcellAI was the lead institution on the PyTDC ICML '25 publication, and the poster presentation was supported by a conference grant. Read our ICML '25 Paper on the underlying technology →

What the platform delivers

  • Interactive AI Agent: An autonomous data scientist and engineer that investigates complex operational questions, uses approved tools, and returns traceable answers.
  • Federated Data Integration: Connects structured data, technical documents, models, and sensor telemetry without forcing expensive replacement migrations.
  • Mission Context Graph: Resolves real-time relationships among systems, telemetry, operations, decisions, and time.
  • Provenance and Traceability: Automatically versions datasets, transformation lineage, and reasoning steps in an auditable graph.
  • Human-Governed Execution: Allows authorized human operators to inspect, approve, redirect, or override agent actions at critical control gates.

How it works

  • Planner → Executor → Critic architecture composes multi-step workflows from standardized tools, secure sandboxes, and enterprise databases.
  • Every run strengthens your provenance graph and semantic layer, improving future planning and ensuring consistent reporting across team-wide workspaces.
  • Model-Agnostic Orchestration: Supports the deployment of specialized local models and deterministic tool sets within a single, highly governed agent workflow.

Product & technical breakthroughs

We provide the missing infrastructure layer for trusted AI-driven engineering. Our platform combines robust agentic reasoning with strict data engineering to solve the trust and validation problem.

  • Semantic Data Layer with Provenance Graph: Based on our ICML 2025 publication, this is a living knowledge graph that tracks every transformation from raw telemetry ingestion to final recommendation. It encodes domain-specific ontologies, allowing agents to understand systems-engineering semantics instead of just raw database schemas.
  • Planner-Executor-Critic Architecture: A robust agentic loop that plans multi-step systems-engineering workflows, executes code in secure sandboxes, and critiques outputs against domain constraints. This prevents the hallucination and fragility common in general-purpose LLM coding tools.
  • Automated Provenance Tracking: Every step of ingestion, transformation, and modeling is automatically versioned. This creates an audit trail essential for regulatory compliance and mission auditability, without requiring manual logging.
  • System and Hardware Integration: We don't just process static files. ArcellAI integrates with operational databases, enterprise data lakes (Snowflake, Databricks), technical manuals, and physical hardware telemetry to unify fragmented operational ecosystems.

Market opportunity

We are capitalizing on a massive structural gap in the engineering and defense market, where systems complexity is outpacing human analytical capacity.

  • TAM: $500B (AI in R&D and Enterprise Systems)
  • SAM: $250B (AI in physical sciences, logistics, and engineering)
  • Primary Beachhead: Defense logistics, maintenance, and systems-engineering operations.
  • Secondary Beachhead: Physical AI, robotics, and complex manufacturing networks.

Our actual moat

Competitors can copy features, but they cannot replicate the proprietary knowledge graph of transformations, system ontologies, and operational context built by our system over thousands of runs.

  • The Provenance Moat: Every pipeline run strengthens our graph, locks in users, and raises switching costs. The more you use ArcellAI, the more it "knows" your specific technical context—making it harder to leave than to stay.
  • The Integration Moat: Deep ties into enterprise databases, technical manuals, and physical systems create high switching costs. We don't just read spreadsheets; we integrate with the operational engines that power your enterprise.
  • The Data Flywheel: More runs → smarter agents → better pipelines → more users → more context. We are building the definitive dataset of how complex operations are run.

Use cases

Defense Logistics & Supply Networks: Connect supplier databases, logistics logs, and operational telemetry to support resilient planning and dependency analysis.

Mission and Engineering Systems: Trace relationships across technical documents, enterprise systems, requirements, and operational events.

Physical AI and Autonomous Systems: Extend governed context and provenance into robotics, sensing, maintenance, simulation, and other physical-system workflows.

Advanced Manufacturing: Sensor fusion, yield analysis, root-cause studies, and adaptive pipeline orchestration with governed experimentation.

Product: ArcellAI Agent and Context Harness

  • Built on peer-reviewed research presented at ICML '25, with domain-specific reasoning via in-context learning and standardized tools/APIs.
  • Autonomous, reproducible workflows and secure integration with enterprise and operational database stacks.

Technical Stack

Our architecture is built to handle the rigorous demands of enterprise engineering, from massive datasets to secure, reproducible compute.

Foundation Models

We leverage state-of-the-art LLMs for reasoning and code generation. Our system is model-agnostic, allowing us to swap in the best model for the task (e.g., Claude 3.5 Sonnet for complex planning, GPT-4o for data synthesis) or run specialized open-source models in secure local environments.

Agent Framework

Our proprietary Planner-Executor-Critic framework orchestrates multi-step workflows.

  • Planner: Decomposes high-level objectives into executable query and code steps.
  • Executor: Runs code in isolated sandboxes with access to authorized tools.
  • Critic: Validates outputs against systems constraints (e.g., statistical significance, schema compliance).

Data Infrastructure

  • Semantic Layer: A knowledge graph that maps raw database schemas to technical ontologies.
  • Provenance Engine: Automatically tracks data lineage, transformation logic, and execution metadata for every run.
  • Hybrid Storage: Optimized for high-throughput operational data (Parquet/Arrow) and complex metadata queries (Graph/Relational).

Engineering Computing

We integrate deeply with the modern data stack and domain-specific analytical libraries. All computation occurs in containerized, isolated environments to ensure reproducibility across different infrastructure.

Deployment

ArcellAI is designed for highly secure environments. It can be deployed as a managed SaaS or entirely within a customer's secure Virtual Private Cloud (VPC) / on-premise infrastructure for maximum security compliance.

Development Resources & Requirements

To build and extend the ArcellAI platform, we rely on a modern, robust tech stack.

Core Requirements

  • Python 3.10+: The backbone of our backend and agent logic.
  • Docker/Kubernetes: For containerization and orchestration of agent sandboxes.
  • PostgreSQL + pgvector: For relational data and vector embeddings.
  • Next.js / React: For our conversational interface and dashboards.

Key Libraries

  • LangChain / LangGraph: For agent orchestration and state management.
  • FastAPI: High-performance API framework for our backend services.

Access & Documentation

  • API Documentation: Comprehensive guides for integrating ArcellAI into your existing enterprise data architectures.
  • SDK: Python and TypeScript SDKs for programmatic access to agent workflows.
  • Community: Join our developer network for support and to connect with other systems engineers.

Why now

Agentic systems unlock the last mile of AI for enterprise engineering and defense by automating the context bottleneck with governance built-in. As organizations move from pilots to production, the winners will have an agentic data layer with provenance, semantics, and reproducibility at the core.

Team

We are an all-MIT CS founding team with deep expertise in AI research and large-scale data infrastructure.

  • Alex Velez-Arce (Founder & CEO/CTO): AI research & strategy at FAANG+, MIT CS, Harvard BioAI. Formerly SWE at Pinterest and AI researcher at Harvard. Built data & ML products accounting for $100M+ in revenue. Won a conference grant for PyTDC's poster presentation at ICML, where ArcellAI was the lead institution.

Our work is peer-reviewed and published at top AI venues including ICML and NeurIPS.

Contact

To request a defense briefing or discuss custom pilot integrations for your operational workflows, reach out at kaela@arcell.ai.