Freelance

A multi-client agentic chat engine with RAG, external tools and generative BI, isolated per client.

The problem

Give each client an assistant that answers any question about their documents and data, connects to their tools and generates charts and dashboards, while keeping data isolated.

Stack decision

FastAPI + LangChain/LangGraph, RAG indexing with pgvector and Cohere Rerank, orchestration of several LLMs by quality/price (OpenRouter, Gemini, Deepseek) and tools via MCP/FastMCP. Fully traceable with Langfuse.

The technical challenge

During indexing, automatically clustering information, detecting discrepancies and allowing manual curation; and isolating data by project_id in a reusable multi-tenant engine.

Impact

A reusable engine in production for several clients (financial Q&A, margin and profitability monitoring with charts).

A generic document-processing engine that, from an Outlook add-in, extracts the order from each email and validates it against SAP in a split document/data screen.

The problem

Turning orders that arrive by email —heterogeneous documents, a different layout per client— into structured orders validated against SAP, solving that the products in the document match neither the name nor the format of those in the client’s database.

Stack decision

A generic document-processing engine combining pdfplumber for structured PDFs, Azure Document AI for complex layouts and an OpenAI LLM for the unstructured parts. Extraction is materialized as a per-client "mask" that is reused on every later order of the same layout, avoiding reprocessing. The Outlook add-in triggers the flow from the email itself, and a split screen sets the attached document against the extracted data —editable and wired to SAP— with automatic price calculation to sanity-check the order.

The technical challenge

Building the generic reusable extraction-mask engine and, on top of it, a learning engine: any user correction (product name, format, SAP mapping) is detected automatically and proposed as a learning to validate. Even a product’s correct format was inferred from the price the client itself set in the order, given that each client had its own pricing.

Impact

Processes around 600 orders a month straight from Outlook. By reusing each client’s mask, recurring orders are resolved without reprocessing, and the learning engine improves extraction with every user-validated correction.

A platform for at-home therapists with a slot optimizer that minimizes travel and waiting time.

The problem

Cutting travel and waiting time for at-home therapists, bringing several roles (therapists, patients, centers, guardians, admin), clinical records, calendar and payments into a single platform.

Stack decision

Next.js/React + FastAPI/SQLAlchemy on PostgreSQL; Stripe (recurring payments and Connect), Google Calendar (two-way sync) and OpenRouteService for real distances/times.

The technical challenge

The slot optimizer: it scores (0–100) each therapist’s gaps and places every patient in the best one using real locations, penalizing dead waiting time and travel.

The cross-cutting infrastructure that orchestrates, deploys and monitors every service and project.

The problem

Managing multiple services and platforms in a reproducible, well-monitored way, with continuous deployment, backups and alerts.

Stack decision

Dokploy + Docker for orchestration, Traefik as reverse proxy, DNS on Cloudflare, WireGuard VPN, SSH tunnels with socat, daily S3 backups and job monitoring with Uptime Kuma plus centralized observability.