# Services offered by Swarnava Dutta

> AI consulting and delivery services across multi-agent systems, RAG architecture, model fine-tuning, and enterprise AI platforms.

## Multi-Agent Systems

Design multi-agent systems with explicit tool access, memory boundaries, approvals, and specialist handoffs.

- Best suited for: Teams replacing repetitive manual work with planner, reviewer, and executor workflows.
- Typical outcomes: Clear agent roles and handoffs; Human review steps and safety rails; Traceable runs and evaluation checkpoints
- Stack: LangGraph, MCP, Tool calling, Guardrails

## RAG, GraphRAG & Knowledge Retrieval

Combine semantic search with entity relationships when answers depend on both document meaning and connected business data.

- Best suited for: Products and internal copilots that need trusted answers from documents, data systems, and live business context.
- Typical outcomes: Knowledge graph, ingestion, and indexing design; Graph-vector retrieval with reranking and citations; Evaluation loops for relevance and answer quality
- Stack: GraphRAG, Knowledge Graphs, Azure AI Search, RAGAS

## Model Fine-tuning & Optimization

Use fine-tuning, prompt design, dataset curation, and inference tuning to improve domain accuracy, latency, and consistency.

- Best suited for: Teams that already see promise but need better domain accuracy, tone control, or task fit.
- Typical outcomes: Dataset curation and benchmark framing; LoRA or PEFT fine-tuning workflow design; Latency, cost, and quality optimization paths
- Stack: LoRA, PEFT, Benchmarks, Inference tuning

## Enterprise AI Architecture

Design identity, data flow, deployment, monitoring, and ownership for AI systems that must run beyond the pilot.

- Best suited for: Teams moving from an AI experiment to a governed platform with clear operational ownership.
- Typical outcomes: A reference setup for services and data flow; Security, compliance, and deployment patterns; Monitoring, team handoff, and maintainability
- Stack: Azure, Kubernetes, CI/CD, Monitoring

## Engagement models

Baseline rates and delivery estimates are shared during the first scoping call.

- **AI Architecture Audit** (1-2 weeks · Fixed scope): Review an existing AI system across retrieval quality, agent design, evaluation coverage, cost, and failure modes. Deliverable: Scored findings with fixes ordered by risk and effort.
- **Discovery Sprint** (2-3 weeks · Fixed scope): Test the use case, data readiness, and architecture options in a working proof of concept. Deliverable: Reference architecture, PoC, and a delivery estimate.
- **End-to-End Delivery** (4-12 weeks · Milestone-based): Design and ship agent workflows or RAG pipelines with evaluation, deployment, documentation, and team handoff. Deliverable: Production system with monitoring, documentation, and handoff.
- **Fractional AI Architect** (Monthly retainer): Ongoing architecture reviews, technical roadmaps, and delivery oversight for teams without a full-time AI architect. Deliverable: Weekly architecture sessions plus async design reviews.