Tom Piaggio

Autonoma · Co-founder & CTO

Agentic end-to-end testing: an autonomous browser agent runs a product's real user flows on every PR and reports failures with video, logs, screenshots, and code-level evidence from an actual execution.

• Led all engineering, hiring, and product roadmap for a 9-engineer team; set monthly technical objectives directly from business targets.
• Designed the browsing agent from scratch. Navigation paths are derived from source code rather than inferred by the model, so it runs on small models: median cost $0.50 per PR, versus $5-15 for a typical LLM code review, which reports opinions on the diff rather than evidence from running it.
• Took the product from zero to paying customers as four technical founders with no sales team: peaked at ~$550K ARR, up to $50K ACV. Weekly regression pipeline at 98% accuracy for a major Argentine bank's digital brand; deployed at fintechs and scale-ups including Superhuman and Mercor. Named best AI startup for innovation and impact on the financial ecosystem at AIFI-25, Buenos Aires, September 2025.
• Built the growth engine with AI-generated, experiment-driven content and no prior marketing experience: 0 to 5.9M search impressions and 68.8K clicks in 12 months (peak month 2.06M / 24K), driving ~1,400 signups/month.
• Recruited a team that includes engineers who have compiled models to hand-tuned x86 assembly for inference performance; ran ~100 interviews at Autonoma on top of ~100 at Google.

Google · Senior Software Engineer (L5), AI & Data

Embedded with Fortune 500 customers on model best practices and production ML deployments.

• Reddit: designed and ran the largest single pipeline in Vision API history: 350M images in a two-week Dataflow/Beam backfill; Google provisioned dedicated capacity for the load. Engagement grew ~5x from the original scope. Built the downstream IAB taxonomy classifier: USE embeddings + KMeans cluster features, 37 per-category forests with swept F1 thresholds, served on Vertex AI.
• McDonald's: designed distributed data gathering and training across a restaurant fleet: a local model per site, a MiNiFi data path that projected out PII before egress and queued what it could not send, and automatic collection and evaluation of every classification.
• Mercado Libre: migrated thousands of tables from Teradata to BigQuery.
• Government intelligence agency: trained a T5 translation model inside an air-gapped environment.
• Created the org's AI learning path for Googlers; grew to ~500 members worldwide. Awarded Initiatives Champion.
• Conducted ~100 technical interviews for L4-L5 ML and software roles.

Metro Smart Cities · Co-founder & CTO

Won a public tender for an airport-scale CCTV analytics system against multinational incumbents (EXA and others) by proposing an AI-based architecture where competitors bid legacy systems. Delivered in one year.

• ~400 fisheye cameras plus three 30MP bullet cameras (~200 cameras per server), sub-second latency, air-gapped, on two nodes.
• One camera to N viewport streams via a grab/retrieve split, so ingestion scaled with cameras and decode with demand. 24 services, 88 CPU, 133 GB, 2 nodes.
• Spun out of the university AI lab below.

Universidad Austral · Founder, AI Laboratory

Founded one of the first university AI labs in Argentina; ran a paper-reading group when little material existed locally. Research project with a hospital on real-time multi-camera tracking mapped to an abstract 3D space, with batch queries over object journeys: analytics for people flow, akin to product analytics for physical spaces.