FABIAN CONTRERAS
PROJECTS
-
Culprit
Root-cause failure diagnosis for production AI agents. Instead of dumping long traces into LLMs, culprit localizes divergence deterministically through a 21-class failure taxonomy, pgvector contrastive diffing against successful runs, and targeted adjudication. Validated on TRAIL and Who&When benchmarks.
-
GMaps Leads
Automated B2B lead generation pipeline extracting structured business intelligence from Google Maps. Built with SeleniumBase and anti-bot evasions, it extracts verified business details, contact info, ratings, and service areas, validating records with Pydantic and exporting to Excel, CSV, JSON, and SQL.
SKILLS
Core technical capabilities and engineering practices developed through building production systems, agent architectures, and data pipelines.
Multi-Agent Systems
Orchestration, dynamic routing, and collaborative agent architectures.
AI Agent & LLM Evaluation
Agent benchmarks, trajectory analysis, hallucination detection, and statistical testing.
Multimodal AI
Vision-language processing, audio pipelines, and document understanding.
Prompt Engineering & Guardrails
Structured schema outputs, few-shot prompting, and safety boundary enforcement.
Web Scraping
Dynamic SPA rendering, proxy rotation, anti-bot mitigation, and DOM parsing.
ETL Data Pipelines
PostgreSQL, DuckDB, Redis, automated ingestion, schema validation, and SQL pipelines.
Backend Engineering
Python (FastAPI, Flask), Node.js, Java (Spring Boot), REST APIs, and worker queues.
CI-CD & Cloud Deployment
Docker, Google Cloud Run, GitHub Actions, and AWS S3 storage.
Root Cause Diagnosis
First-principles debugging, isolating core defects, and failure reproduction.
Fault-Tolerant Pipelines
Exponential backoff, retry mechanisms, circuit breaking, and failure recovery.
System Observability
Structured logging, error telemetry, audit trails, and runtime metrics.
Benchmark-Driven Development
Standardized scorecards, regression baselines, and quantitative targets.
Automated Regression Testing
Unit, integration, and end-to-end testing suites with CI assertions.
Statistical Significance Testing
Confidence intervals, A/B hypothesis verification, and variance control.
Zero-to-One Prototyping
Rapid proof-of-concept delivery, architectural spikes, and viable solutions.
ABOUT ME
I am an AI Systems Engineer and Data Scientist specializing in agent architectures, LLM evaluation infrastructure, and end-to-end automation. With a solid foundation in mathematics and machine learning, I focus on building reliable systems: deterministic agent workflows, rigorous statistical regression testing, and resilient data extraction pipelines. I engineer practical, production-ready solutions designed for reliability, measurable performance, and real business value.