Building reliable backend products and AI infrastructure with an emphasis on evidence, provenance, controlled execution, independent acceptance, and operational continuity.
C# / .NET · Python · PostgreSQL · FastAPI · RAG · AI Systems · Evaluation
Backend and AI systems engineer working across product engineering, RAG, agent infrastructure, distributed execution, and evidence-driven evaluation.
My work focuses on systems where correctness cannot be inferred from implementation alone. I design explicit contracts for ownership, execution, verification, acceptance, recovery, and external side effects, then use tests and runtime evidence to establish what is actually demonstrated.
A recurring rule across the projects is:
IMPLEMENTED != EXECUTED
EXECUTED != VERIFIED
VERIFIED != ACCEPTED
ACCEPTED != PROMOTED
- Backend & product systems — APIs, domain boundaries, PostgreSQL, queues, concurrency, authentication, idempotency, integrations, CI/CD and operational tooling.
- RAG & knowledge systems — hybrid retrieval, provenance, privacy boundaries, canonical identity and reproducible ingestion.
- AI-agent infrastructure — runtime admission, qualification, supervision, policy/budget boundaries, durable state, fencing, recovery and independent acceptance.
- Evidence-driven engineering — controlled comparisons, falsifiable hypotheses, benchmark fidelity, explicit failure classes and claims limited to executed evidence.
|
Automotive marketplace + knowledge platform Two bounded systems working together: Podium7 produces reconciled automotive knowledge and provenance; BPT2 consumes versioned contracts into its canonical catalog and marketplace. BPT2 engineering
Podium7 integration
Status: BPT2 post-MVP operational baseline · hosted staging acceptance is the current evidence boundary · private source |
🧪 CodeProEvidence-driven software-agent chassis Executor-agnostic engineering chassis for testing software-agent mechanisms through explicit contracts, falsifiable hypotheses and reproducible evidence. Engineering focus
Status: active empirical chassis · implementation gate for the shootout runner is closed · real-task executor validation remains the next evidence boundary |
⚖️ RpyRAG backend for legal-process analysis Backend system built with FastAPI, PostgreSQL 16 and pgvector, designed around tenant isolation, provenance, concurrency, privacy and reproducible operations. Engineering focus
Status: offline path qualified and reproducible |
🧠 MetaOMeta-orchestrator / AI control plane Framework-neutral control plane for selecting, governing, supervising and independently accepting work from pluggable agent orchestrators. Engineering focus
Status: advanced operational baseline · 23-gate Project Plane release battery recorded · integrated hosted validation remains an evidence boundary |
| Project | What it demonstrates |
|---|---|
| RJ | .NET 10 / C# 14, ASP.NET Core, modular monolith boundaries, architecture tests, deterministic build and legal RAG work. |
| SMAG | Installable governance layer for coding executors: permissions, isolated work, explicit outcomes, evidence and acceptance. Private source. |
| NDV | Research program on minimum-sufficient capability composition and the verified-success / cost / latency frontier. Current engineering mode: REUSE_ONLY; no NDV runtime architecture is approved. |
| NaIA | User-facing governed capability platform: intent → plan → policy/approval → tools/providers → evidence → resumable state. Includes live-provider adapters, while live validation remains credential/environment scoped. Private source. |
| MEDVI | Full-stack health product with Next.js, PostgreSQL/Prisma, auth, Stripe checkout/webhook recovery, rate limiting, CSP and staged operational rollout. Production promotion remains unauthorized pending remaining environment gates. Private source. |
I keep architecture subordinate to evidence rather than the other way around:
problem
↓
explicit hypothesis / invariant
↓
smallest useful implementation
↓
execution
↓
verification
↓
evidence
↓
keep / change / remove
For AI systems, executor output is treated as an observation, not as acceptance authority. For conventional backend systems, the same discipline appears in ownership boundaries, idempotency, concurrency tests, migration gates, security controls and explicit external-side-effect semantics.
BackendC# .NET ASP.NET CoreABPPython FastAPITypeScript Next.js
|
DataPostgreSQLpgvectorEF Core PrismaSQL
|
AI EngineeringRAG LLMsAgent SystemsGovernanceEvaluation Benchmarks
|
PlatformDockerGitHub ActionsCI/CDLinux Git
|
Backend Engineering · AI Systems · RAG · Agent Infrastructure · Distributed Systems · Evaluation · Reliability · Operational Continuity