Pantera AI is the company I am currently co-founding, where I serve as Chief Scientist and build the platform end-to-end myself. That includes the AI systems, product architecture, browser automation, backend, infrastructure, and production execution, while my co-founders handle the other sides of the company. We are building an automation system that learns browser and business workflows from recordings, then replays them deterministically at scale for teams that cannot tolerate flaky execution or black-box behavior.
Challenge
A surprising amount of real business work still happens in browser UIs, internal portals, PDFs, spreadsheets, and semi-structured operational processes. Traditional automation is too rigid for these workflows, while pure LLM agents are too unreliable for environments where correctness, repeatability, and auditability matter.
That gap is exactly what Pantera AI is designed to solve.
What We Built
Pantera AI turns recorded workflows into repeatable automations that can execute reliably in production environments. The system combines AI-based interpretation with explicit execution logic, validations, and recovery mechanisms so workflows can keep running even when inputs vary or interfaces drift.
The first concrete wedge is regulated operational work, starting with Mexican tax compliance, where mistakes are expensive and manual repetition burns significant time.
My Role
As Co-Founder and Chief Scientist, I am building the platform end-to-end: AI systems, workflow orchestration, browser automation, backend services, infrastructure, and production reliability.
My work spans:
- AI system architecture
- Workflow representation and orchestration
- LLM-assisted interpretation of recorded business processes
- Deterministic execution design
- Recovery and self-healing mechanisms
- Backend and infrastructure
- Technical product direction for a compliance-heavy domain
Core System Design
- Recorded workflows become structured automation logic instead of remaining one-off demos
- Browser and business actions are replayed deterministically rather than delegated to unconstrained agent behavior
- Validation layers reduce silent failures and catch bad state before execution continues
- Self-healing mechanisms help the system recover from UI drift, transient issues, and real-world edge cases
- Outputs remain traceable enough for teams that need confidence, reviewability, and operational trust
Why It Matters
Most AI automation products still behave like black boxes: impressive when they work, painful when they do not. Pantera AI is aimed at the harder category of automation where teams need repeatability, explicit logic, and the ability to trust execution in production.
That is the part of AI automation I find most interesting: not generating one good answer, but building systems that can reliably do useful work over and over again.