Frequently Asked Technical & Commercial Questions
Complete technical answer directory for enterprise software engineering buyers. Explore 20 detailed answers covering agentic architecture, LLM data privacy, IP ownership, delivery timelines, pricing models, and EU AI Act compliance.
Engagement & Project Process (4 Questions)
What is the typical delivery timeline for a custom multi-agent system? ↓
Standard production multi-agent engagements require 8 to 14 weeks from initial architecture design to validated deployment.
How does Esaholic handle team onboarding and project kickoff? ↓
We assign a dedicated lead engineer within 5 business days following feasibility audit completion and NDA signing.
Do you offer ongoing post-deployment maintenance and monitoring? ↓
Yes, we provide tiered MLOps SLAs covering vector index re-indexing, drift monitoring, and model update pipelines.
Can your team work alongside our internal engineering developers? ↓
Yes, we frequently operate as an embedded specialist AI architecture squad integrated into your existing Git workflows.
Technical Architecture & Stack (4 Questions)
What frameworks do you use for autonomous agent orchestration? ↓
We build deterministic state graphs using LangGraph, AutoGen, and custom Python async event loops paired with MCP servers.
How do you guarantee vector retrieval accuracy and eliminate hallucinations? ↓
We enforce hybrid dense-sparse vector search, Reranking models (Cohere/BGE), knowledge graph validation, and strict Pydantic schemas.
Can you deploy models on-premise or within isolated VPC private clouds? ↓
Yes, we regularly deploy quantized open-weights models (Llama 3.3, DeepSeek) inside client AWS VPC, Azure GovCloud, or on-prem hardware.
What latency targets can your agentic architectures achieve? ↓
Our multi-agent routing engines achieve sub-350ms median latency through parallel tool execution and small model distillation.
Commercial Models & Budgeting (4 Questions)
What are your standard pricing models? ↓
We offer fixed-scope milestone deliverables ($25k-$150k) and dedicated monthly engineering team retainers ($18k-$45k/mo).
Are LLM API consumption costs included in project estimates? ↓
No, model provider API costs (OpenAI, Anthropic, AWS Bedrock) are billed directly to your corporate cloud accounts.
Do you charge royalties or recurring software license fees? ↓
No. All code and custom architectures created during the engagement are transferred to you without ongoing license fees.
What is the minimum budget requirement to engage Esaholic? ↓
Our minimum production project scope starts at $25,000 for proof-of-feasibility audit and baseline model pipeline builds.
Data Privacy, Security & IP Rights (4 Questions)
Who owns the intellectual property and code developed during the project? ↓
You retain 100% full legal ownership of all custom source code, trained model weights, vector schemas, and documentation.
Is our proprietary enterprise data ever used to train public LLM models? ↓
Never. We strictly enforce zero-data-retention APIs and self-hosted open models ensuring your data never trains vendor models.
Are your development environments SOC 2 Type II compliant? ↓
Yes, Esaholic maintains SOC 2 Type II compliance and undergoes annual independent third-party security audits.
How do you handle sensitive customer PII before passing data to LLMs? ↓
We implement automated local PII masking and tokenization layers before any payload crosses network boundaries.
Compliance & AI Ethics (4 Questions)
Does your AI code comply with the European Union AI Act? ↓
Yes, we classify system risk tiers and implement required transparency logs, human oversight triggers, and audit trails.
Are your AI engineering methodologies ISO 42001 certified? ↓
Yes, our internal AI Management System adheres to ISO 42001 standards for model safety, fairness, and risk governance.
How do you test AI agents against prompt injection and jailbreak attacks? ↓
Our dedicated security squad executes automated red-teaming test suites using OWASP Top 10 for LLMs benchmarks.
What audit logging is provided for autonomous agent decisions? ↓
We record full deterministic state histories in structured JSON logs including agent tool calls, prompts, and validator results.