ENGINEERING JOURNAL

Architectural Insights & Systems Research

Authored by OXELVO engineering research. Rigorous technical analyses, benchmark breakdowns, and pragmatic engineering principles directly from production trenches. Zero marketing fluff.

AI ARCHITECTURE•October 12, 2025•8 min read

ADOXELVO Engineering

Deterministic Agentic Workflows: Replacing Prompts with State Machines in Production

Why relying purely on LLM reasoning chains inevitably fails in commercial production. How to combine LangGraph state machines, Pydantic guardrails, and idempotent execution loops to build systems with 99.9% reliability.

KEY ARCHITECTURAL TAKEAWAYS

Unbounded prompt chains suffer catastrophic error compounding after step 4
Typed schema validation (Pydantic/Zod) acts as a strict compile-time firewall
State machines preserve execution context across process crashes and reboots
Human-in-the-loop approval gates are mandatory for non-idempotent operations

PRODUCTION SPECIFICATIONpython

@dataclass
class WorkflowState:
    task_id: str
    validated_payload: Optional[Dict] = None
    retry_count: int = 0
    status: Literal["init", "evaluating", "committed", "failed"] = "init"

def validation_node(state: WorkflowState) -> WorkflowState:
    try:
        # Strict Pydantic schema validation firewall
        validated = InvoiceSchema.model_validate(state.validated_payload)
        return replace(state, status="committed", validated_payload=validated.model_dump())
    except ValidationError as err:
        return replace(state, status="failed", retry_count=state.retry_count + 1)

Read Full Technical ArticleOXELVO RESEARCH

SEARCH & RETRIEVAL•September 28, 2025•11 min read

ADOXELVO Engineering

Sub-150ms RAG Pipelines: Hybrid Vector Search & Re-Ranking at Scale

A deep dive into our production RAG stack. We analyze the latency and precision tradeoffs between dense cosine embeddings and BM25 sparse keyword indices, and how cross-encoder reranking eliminates data hallucination.

KEY ARCHITECTURAL TAKEAWAYS

Dense vectors alone miss exact keyword matches like SKUs and transaction IDs
Reciprocal Rank Fusion (RRF) delivers 23% higher precision on technical corpora
Semantic caching in Redis resolves 44% of repetitive corporate queries in < 12ms
Cross-encoder reranking eliminates irrelevant context before prompt assembly

PRODUCTION SPECIFICATIONpython

def reciprocal_rank_fusion(dense_ranks: dict, sparse_ranks: dict, k: int = 60):
    scores = defaultdict(float)
    for doc, rank in dense_ranks.items():
        scores[doc] += 1.0 / (k + rank)
    for doc, rank in sparse_ranks.items():
        scores[doc] += 1.0 / (k + rank)
    return sorted(scores.items(), key=lambda x: x[1], reverse=True)

Read Full Technical ArticleOXELVO RESEARCH

DATABASE & SCALE•August 15, 2025•9 min read

ADOXELVO Engineering

Architecting Multi-Tenant SaaS with PostgreSQL Row Level Security

Stop leaking tenant data with application-level WHERE tenant_id clauses. Learn how PostgreSQL Row Level Security (RLS) enforces cryptographic tenant boundaries directly at the database engine kernel.

KEY ARCHITECTURAL TAKEAWAYS

Application bugs cannot bypass RLS when executed in authenticated sessions
Setting app.current_tenant via connection pool wrappers maintains zero overhead
Simplifies SOC2 and ISO tenant isolation compliance audits dramatically
Works seamlessly with modern TypeScript ORMs like Prisma and Drizzle

PRODUCTION SPECIFICATIONsql

-- Enable Row Level Security on the table
ALTER TABLE enterprise_records ENABLE ROW LEVEL SECURITY;

-- Define tenant isolation policy for all operations
CREATE POLICY tenant_isolation_policy ON enterprise_records
FOR ALL
TO authenticated_app_user
USING (tenant_id = NULLIF(current_setting('app.current_tenant', true), '')::uuid);

Read Full Technical ArticleOXELVO RESEARCH

DEVSECOPS & QA•July 20, 2025•7 min read

ADOXELVO Engineering

Evaluating LLM Drift and Regression in Continuous Integration

How to set up automated model evaluation pipelines in GitHub Actions. We run synthetic gold-standard datasets against every PR to catch prompt degradation, tone shifts, and tool schema breaks before deployment.

KEY ARCHITECTURAL TAKEAWAYS

LLM provider model updates can silently alter reasoning behavior and output format
CI test suites must evaluate both semantic accuracy and JSON schema conformity
Threshold gating prevents broken agents from being merged into main branch
Synthetic test datasets must mirror real production edge cases

PRODUCTION SPECIFICATIONyaml

steps:
  - name: Run DeepEval Model Regression Suite
    run: pytest tests/evals/test_agent_reasoning.py --threshold=0.95
    env:
      EVAL_API_KEY: ${{ secrets.EVAL_API_KEY }}

Read Full Technical ArticleOXELVO RESEARCH

Have a question about these architectural benchmarks?

OXELVO Solutions Architects are available for architectural consultations on your specific system topology.

Schedule Architectural Consultation