I build production systems clients own and run in their own cloud — RAG over operational data, agent workflows, custom apps. The work that pulls me in is the cross-system flows their legacy ERP, BI, and CRM stack couldn't deliver: reporting, approvals, and exports the per-seat tools were never going to build. The licenses retire alongside.
The repos below are the public ones; embedded engagements happen privately.
Practice (outside builds): evanparra.ai
Founding engagement — Regional commercial electrical contractor (NE Florida, anonymized). Embedded in the client's Azure tenant, replacing per-seat ERP workflows app by app — cost-to-complete forecasting, change orders, approvals, and the cross-system flows the per-seat tools were never going to build. ~$150K/yr in retired per-seat ERP cost (and growing). The ERP stays the system of record; the seats don't.
Cross-vertical proof — my own products, both live:
- TextTimeline — legal document intelligence. Messy text exports become attorney-ready chronological timelines with 100% source citations. FAISS + BM25 hybrid retrieval, Cloud Run, Firestore, Gemini. (Source private — paid product.)
- GammaRips — autonomous overnight options-flow scanner. 14 Cloud Run services, ~20 schedulers, multi-agent ADK publishing layer with deterministic compliance gating.
| Area | Focus |
|---|---|
| Custom workflow apps | Production systems in client clouds — cross-system reporting, multi-stage approvals, and exports legacy ERP, BI, and CRM tools couldn't build; per-seat licenses retire alongside |
| Generative AI | Diffusion models, fine-tuning (LoRA/QLoRA), multi-modal pipelines, content safety |
| LLM Applications | RAG systems, prompt chaining, MCP tool servers, agent orchestration |
| ML Pipelines | End-to-end data ingestion → feature engineering → model deployment |
| Evaluation & Safety | Hallucination detection, factual accuracy, brand safety, A/B benchmarking |
| MLOps | CI/CD for ML, model versioning, monitoring, cost optimization |
| Data Engineering | BigQuery, ETL/ELT pipelines, multi-source integration |
Autonomous trading signal platform processing ~10GB daily market data. Full MLOps lifecycle from ingestion to deployment.
- LLM-augmented ETL with prompt chaining
- MCP server for AI agent tool-calling
- CI/CD: GitHub Actions → Cloud Build → Cloud Run
- 50% inference cost reduction via dynamic model routing
Stack: Python, BigQuery, Vertex AI, Cloud Run, Pub/Sub, MCP
Customer-facing surface for GammaRips. Daily mechanically-held picks, subscription billing, compliance disclosures.
ML core for the GammaRips signal stack. ~3x precision lift versus baseline, with a quarterly retraining cadence.
Model Context Protocol server enabling AI agents to query real-time financial data. Production-deployed on Cloud Run with SSE transport.
Stack: Python, FastMCP, BigQuery, Cloud Run
Production evaluation framework for generative AI systems. NLI-based hallucination detection, factual accuracy verification, content safety scoring, and A/B model benchmarking with statistical significance testing.
- Hallucination detection via cross-encoder NLI + semantic similarity
- Brand safety scoring with configurable content rating (G/PG/PG-13/R)
- A/B comparison engine with paired t-test and effect size analysis
- HTML + JSON reporting for CI/CD integration
Stack: Transformers, Sentence-Transformers, Detoxify, Scikit-Learn, Pydantic
Parameter-efficient fine-tuning of LLMs using QLoRA. 4-bit quantization with PEFT adapters, full training pipeline with experiment tracking.
- QLoRA with BitsAndBytes NF4 quantization
- SFTTrainer from TRL with gradient accumulation
- Weights & Biases experiment tracking and evaluation
- Interactive inference with streaming output
Stack: Transformers, PEFT, TRL, Accelerate, BitsAndBytes, W&B
Text-to-image generation with Stable Diffusion XL, IP-Adapter style conditioning, and content safety guardrails.
- SDXL base + refiner pipeline with safety-first architecture
- Brand consistency scoring via CLIP embeddings
- Content rating system (G/PG/PG-13) for family-friendly generation
- NSFW classification and automated content filtering
Stack: Diffusers, Transformers, OpenCLIP, PyTorch, Pillow
Cross-modal AI pipeline: audio transcription → LLM analysis → structured output. Dual backend support with async orchestration.
- Whisper + Google Cloud Speech-to-Text dual backends
- Gemini-powered analysis: sentiment, entities, topics, action items
- Pydantic-validated structured JSON output
- Async pipeline with retry logic and batch processing
Stack: OpenAI Whisper, Google Generative AI, Pydantic, PyDub
Knowledge-grounded clinical Q&A agent using Graph RAG with Google Cloud. Combines medical knowledge graphs with retrieval-augmented generation for accurate, citation-backed healthcare answers.
Stack: Python, ADK, Gemini, Spanner Graph, Vertex AI, Cloud Run
Multi-agent system automating invoice lifecycle: Ingestion → Validation → Approval → Payment. Self-correction loops for data extraction.
Stack: Python, LangGraph, xAI Grok, FastAPI, Cloud Run
Secure file storage with user isolation and irreversible PII redaction using event-driven architecture.
Stack: Cloud Run, Cloud DLP, Vertex AI, FastAPI
Multi-document scientific paper Q&A with citation tracking. Vertex AI Vector Search + Gemini.
Stack: RAG, Vertex AI, Gemini, FastAPI, Firestore
Computer vision research from M.S. AI coursework at Florida Atlantic University — end-to-end guide for fine-tuning YOLOv9 on custom datasets.
Stack: PyTorch, YOLO, Computer Vision
Generative AI: Diffusers, PEFT/LoRA, Whisper, Stable Diffusion, CLIP
ML/AI: Vertex AI, Gemini, TensorFlow, PyTorch, Scikit-Learn
Evaluation: Sentence-Transformers, Detoxify, W&B, custom frameworks
Cloud: GCP (BigQuery, Cloud Run, Pub/Sub, Cloud Functions, Vertex AI), Azure (client engagements)
MLOps: GitHub Actions, Cloud Build, Docker, Model Registry
Data: Python, SQL, Pandas, dbt, Airflow
Backend: FastAPI, Python, Node.js
Frontend: Next.js, React, TypeScript
- M.S. Artificial Intelligence — Florida Atlantic University
- B.A. Economics — Florida International University
- Google Professional Machine Learning Engineer
- Google Advanced Data Analytics
- Lean Six Sigma Green Belt
- EVANPARRA.AI LLC — SAM.gov registered (UEI FPLQTQK39ZE1), SBIR/STTR eligible (CLARA / DARPA proposal submitted Mar 2026)
- Practice: evanparra.ai
- Email: evan@evanparra.ai
- LinkedIn: linkedin.com/in/evanparra
Booking discovery engagements through evanparra.ai. Based in NE Florida; embedded work in Azure or GCP tenants.
5. Update pinned repos (§3)
6. Fix gammarips-engine homepage URL + add lora-finetune-lab topics (§4a-b)