Selected Work
A decade of engagements across regulated, high-stakes domains — and open-source you can verify yourself. Every number below is from real production work.
Verified outcomes
Investing-media data platform
Leading Python API & microservices work across a fleet of production Django/FastAPI services at a national investing-media platform, FJS targeted the parts of the system that were too slow or too fragile:
- 75% latency reduction on the market-data pipeline (1.2s → 300ms) by replacing sequential fetches with a concurrent fan-out and an enlarged connection pool, plus partial-results fallback so a single timeout no longer fails the whole request.
- 17× faster portfolio watchlist endpoint (9s → 530ms) by eliminating redundant pipeline re-execution.
- 99.9% fewer queries and 60× less database lock time (3,960s → 66s) on a bulk financial-data ingestion pipeline rewritten from row-by-row writes into a staged bulk pipeline.
- Built an event-driven subscription-lifecycle layer on AWS EventBridge/SQS, replacing brittle logic inherited from a legacy service.
Open source — proof you can check yourself
Anonymized client metrics only go so far. Our open-source libraries are public, named, and run in production by other teams:
- TextSpitter — multi-format document text extraction (PDF, DOCX, and 50+ formats), 70+ tests across a Python 3.12–3.14 CI matrix, published on PyPI.
- SQLModel CRUD Utilities — standardized async/sync CRUD on top of SQLModel/SQLAlchemy, with transactions, soft-delete, and audit mixins, published on PyPI.
- More at github.com/fsecada01.
In-house R&D — autonomous performance engineering
We invest in our own tooling, including an autonomous agent that profiles CPU-bound Python and C++ hot paths, rewrites the worst offenders in Rust, benchmarks before and after, and keeps only the changes that pass the existing test suite and clear a measured speedup threshold — reverting anything that doesn't. Its AI layer is deliberately vendor-agnostic and local-first: open-weight models running on our own hardware by default, frontier APIs only when they earn it. It's the same discipline we bring to client systems — measured results, correctness gates, and no dependence on a single AI provider.
Healthcare — a major hospital system & Medicaid waiver automation
At a major hospital system's diagnostics & AI analytics group, FJS owned the migration of a legacy Django monolith to FastAPI microservices on Azure — 50% lower cold-start latency, 60% faster clinical-data queries, and a 60% jump in test coverage, all inside HIPAA/HITECH guardrails.
Independently, FJS builds Formana, an automation platform for New York State's Medicaid waiver programs (TBI & NHTD): Django with LLM-backed document generation, with role-based access control, granular PHI permissions, and end-to-end audit trails.
Financial services & cybersecurity — a global bank
Over a multi-year engagement with a global bank, FJS helped re-architect a worldwide cybersecurity control platform from a Django monolith into containerized microservices on AWS ECS — 50% faster deployments, 30% fewer production incidents, and regulatory-reporting dashboards processing millions of transactions a day, plus mentoring and team growth.
Independent product work
Through FJS we've delivered 15+ full-stack builds for SMBs, startups, nonprofits, and boutique investment firms — from early-stage MVPs to production platforms.
How an engagement actually goes
Numbers are the result; the work is the reasoning that gets you there. One representative engagement:
The scenario
A high-traffic investing-media platform had a report-generation endpoint that took roughly nine seconds to respond. Under load it tied up database connections and dragged the rest of the application down with it — the kind of slow that costs readers and revenue.
The approach
Rather than reach for more servers, we profiled the request to find where the time and the contention actually were: redundant queries, a hot code path doing work on every call, and locks held far longer than they needed to be. We fixed the bottleneck itself — restructuring the queries and the locking so the database spent its time serving requests instead of waiting on them.
The outcome
The endpoint went from about nine seconds to 530 milliseconds — a 75% cut in latency and roughly 60× less time spent holding database locks. No new infrastructure, and a measurement to prove every claim.