Reference Implementations & Architectures
At Botmartz, we believe engineering capability is proven through transparent code, benchmarked architectures, and reproducible reference builds—not fabricated marketing metrics. Explore how our production systems are architected.
VisaDesk — Visa & Migration Consultancy Operations OS
An enterprise-grade modular monolith engineered with FastAPI, PostgreSQL 16 + pgvector, Redis, and Next.js 15. Features a complete Hybrid Search (BM25 + Dense) RAG pipeline with Corrective RAG (CRAG) system guardrails and multi-role RBAC.
Architectural Hallmarks
- •Hybrid retrieval combining dense vector similarity with sparse BM25 keyword matching
- •Cross-encoder reranking postprocessors (BGE / Cohere)
- •Strict sufficiency gates that enforce refusal when context is missing
- •Multi-turn conversational memory persisted via Redis
Tech Stack & Tooling
Deterministic Multi-Actor Customer Operations State Machine
A constrained AI agent workflow built on LangGraph that isolates deterministic business logic from probabilistic LLM generation. Incorporates human-in-the-loop validation for any state-changing API invocation.
Architectural Hallmarks
- •Explicit cyclic state machines with conditional edge validation
- •Human-in-the-loop approval nodes for write operations
- •Tool-calling guardrails with strict Pydantic JSON schema verification
- •Complete execution tracing and latency telemetry
Tech Stack & Tooling
Healthcare Payer Policy Ingestion & RAGAS Evaluation Suite
An open, reproducible reference benchmark over complex clinical policy documents and tables. Features tiered document parsing (fast born-digital path vs. layout OCR), bounding-box provenance tracing, and automated RAGAS faithfulness testing.
Architectural Hallmarks
- •Content-addressed chunk storage with page-level bounding box provenance
- •Reciprocal Rank Fusion (RRF) merging dense and sparse indexes
- •Automated evaluation harness measuring Context Recall, Precision, and Faithfulness
- •Designed for 100% on-premise deployment with zero external API data egress
Tech Stack & Tooling
Want to see how these architectures apply to your documents?
Book a 30-minute technical discovery call with our founder & GenAI architect to review your corpus and determine RAG feasibility.