Agentic AI Research & Advisory

Making agentic AI
work in the real world

Theseus Research works at the intersection of open-source and frontier AI — building tools, shipping systems, and advising organizations on how to deploy autonomous agents that reason, plan, and execute in production environments.

Agents should be accessible
to everyone who builds

The agentic AI revolution shouldn't be gated behind billion-dollar labs. We work across the full spectrum — from open-weight models running on local hardware to frontier systems from Anthropic, OpenAI, and Google — because the best agent architecture depends on the problem, not the hype cycle.

01

Open Source First

We build tools in the open. Our research platforms, inference libraries, and agent frameworks are Apache-2.0 licensed because advancing the field requires lowering the barrier to entry.

02

Model Agnostic

Real agentic systems need flexibility. We design architectures that work with Ollama on a laptop and Claude on an enterprise API — because lock-in is the enemy of good engineering.

03

Production Tested

Theory papers are cheap. We focus on agentic systems that survive contact with reality — regulated industries, messy data, real users, and the thousand edge cases that separate demos from deployments.

Tools we've shipped

Research infrastructure and developer tooling — built to be used in production, not just published.

Theseus Insight

An end-to-end research intelligence platform. Fetches, ranks, and analyzes papers from ArXiv and PDFs, generates newsletters and podcasts, and visualizes research landscapes through interactive mind-maps, timelines, and 3D star maps.

Multi-Agent Research System — Dual-mode orchestration with specialized agents for parallel analysis, verification, and synthesis
Unified LLM Interface — Seamless support for OpenAI, Anthropic, Gemini, Ollama, LM Studio, and LlamaCPP through LLMFactory
Hybrid Vector Search — PostgreSQL + pgvector combining semantic similarity with full-text search for superior accuracy
Interactive Visualizations — Mind-map explorer, research timelines, and 3D constellation star maps for up to 10K papers
Multi-Server Inference — Distribute LLM workloads across heterogeneous hardware with per-server model configuration
Newsletter & Podcast Pipeline — Automated content generation from curated research with email distribution and TTS

LLMFactory

A unified inference library that abstracts away provider differences so you can write agent logic once and run it anywhere — from a laptop with Ollama to a production API calling Claude. Born out of the practical need to stop rewriting boilerplate every time you swap a model.

7 Provider Backends — OpenAI, Anthropic, Google Gemini, Ollama, LM Studio, LlamaCPP, and any OpenAI-compatible endpoint through a single consistent API
Streaming Everywhere — Token-by-token streaming that works identically across all providers, local and cloud
Structured Output — Schema-based JSON generation with Pydantic models, with automatic fallback parsing for providers that don't support it natively
Multimodal Support — Vision capabilities for compatible models, same interface regardless of provider
Embedding Models — HuggingFace Sentence Transformers with GPU acceleration and Ollama-based embeddings for local vector generation
Type-Safe & Configurable — Full type hints, flexible configuration via environment variables, direct parameters, or database-stored credentials

Where we can help

Strategic guidance for organizations building with agentic AI — grounded in years of shipping production systems in regulated industries.

Strategy

Agentic AI Architecture

Moving from chatbots to autonomous workflows is an architecture problem, not a model problem. We help teams design agent systems that decompose reliably, handle failure gracefully, and actually ship.

Implementation

Open Source + Frontier Integration

The right model for the job isn't always the most expensive one. We help organizations build flexible inference stacks that leverage both open-weight and proprietary models where each excels.

Operations

AI in Regulated Industries

Deploying AI in pharma, biotech, and other regulated environments requires understanding both the technology and the compliance landscape. We've done it at scale.

Process

Agentic Development Workflows

AI-assisted development is more than autocomplete. We've developed structured methodologies for teams to move from "vibe coding" to disciplined, reproducible agentic development.

Notes from the field

Occasional writing on what's actually working in agentic AI — practical lessons from building and deploying autonomous systems in production, not benchmark commentary. Covering open-source tooling, agent architectures, and the messy reality of AI in regulated industries.

Read on Substack

Background

Theseus Research is led by Christian Merrill, who brings over a decade of experience at the intersection of AI, data science, and regulated industries. He currently leads agentic AI initiatives at a top-5 pharmaceutical company, designing and deploying generative and autonomous AI systems across R&D — from document processing pipelines to multi-agent research workflows.

Prior to industry, Christian served in the U.S. Army with specialized language training. That background in structured analysis and mission-driven problem decomposition informs a distinctive approach to AI system design: always start with the objective, work backwards to the architecture.

Beyond pharma, Christian advises startups applying AI to operational workflows — including an AI-powered home services platform using intelligent diagnostics to transform how residential repairs are estimated and dispatched.

Theseus Research was founded on a simple observation: the gap between what agentic AI can do and what most organizations know how to deploy is enormous — and closing it requires builders who work in the open.

Focus
Agentic AI — Pharma & Life Sciences R&D
Experience
Director-level AI/Data Science across top pharma
Advisory
AI Advisor to an AI-powered home services platform
Education
MBA · MS Information Technology · BS Biochemistry
Background
U.S. Army Veteran

Let's talk

Whether you're exploring agentic AI for your organization, want to contribute to our open-source projects, or just want to talk shop about what's actually working in production — reach out. You can also follow along on Substack.

christian@theseusresearch.ai