Applied AI architecture and engineering collaboration
I am a Senior AI Engineer at Sutra.AI, based in Bhopal. My work spans production AI systems, and I welcome conversations about selected architecture, advisory, and applied AI collaborations.
Where I am most useful
Four areas, each scoped around a concrete problem. If your situation does not fit one of them, say so in the brief — the honest answer is often that someone else is a better fit.
AI system and architecture review
A system is in production, but latency, cost, or evaluation quality has drifted and nobody is certain why.
I review the retrieval, orchestration, and serving path end to end, then give you a written assessment of what is holding it back and what to change first.
What you get- Architecture and data-flow walkthrough
- Prioritised findings with effort and impact
- Concrete remediation plan, not a slide deck
A good fit ifTeams with a working AI feature that is underperforming.
- LangGraph
- LangChain
- FastAPI
- AWS Bedrock
- GCP Vertex AI
RAG and knowledge systems
Internal documents, databases, and catalogues exist, but answers are unreliable, slow, or hard to trust.
I design and build retrieval pipelines with hybrid search, schema-aware SQL routing, reranking, and evaluation so the system can be measured instead of guessed at.
What you get- Retrieval design with chunking and reranking strategy
- Evaluation set and measurable quality baseline
- Working pipeline with the observability to maintain it
A good fit ifTeams with knowledge spread across documents and live databases.
- Qdrant
- Pinecone
- LangChain
- PostgreSQL
- Gemini
- Claude
Agent evaluation and workflow design
Agents demo well but behave unpredictably on real inputs, and manual review absorbs the time they were meant to save.
I build the evaluation harness and the workflow boundaries together: what the agent may do, how its output is checked, and where a human signs off.
What you get- Task-level evaluation harness with regression runs
- Tool-selection and guardrail design
- Automation with explicit human review points
A good fit ifTeams moving an agent from prototype into daily operations.
- LangGraph
- CrewAI
- MCP / FastMCP
- Google ADK
- Python
Production ML and MLOps delivery
Models work in notebooks but are slow to deploy, hard to monitor, or expensive to run.
I containerise and instrument the model path, set up reproducible deployment and tracking, and tune inference cost without trading away accuracy.
What you get- Containerised service with CI/CD pipeline
- Experiment tracking and versioned models
- Cost and latency baseline with tuning notes
A good fit ifTeams with models that need a reliable route to production.
- Docker
- FastAPI
- MLflow
- GitHub Actions
- AWS
- GCP
How I work
Problem first
A useful conversation starts with the problem, the constraints, and what has already been tried. I will tell you early if the work is not a fit.
Written findings
Reviews and recommendations come as written artefacts you can act on with or without me afterwards.
Clear boundaries
Scope, assumptions, and what is out of scope are agreed before work starts, so there are no surprises on either side.
- Sutra.AIEnterprise AI & Agents
- Revive AnalyticsColorado, USA
- Saffire TradewingsQuantitative ML
- Tech Consulting PartnersLondon, UK
- Ineuron IntelligenceBangalore, India
- Rocket Capital AI1st Place Winner (5,000+ competitors)
- Citi Bank HackathonTop 10 National Finalist
- Accenture AI HackathonTop 8 Finalist
- Stanford UniversityMachine Learning
- IBMDeep Learning with PyTorch
Common questions
- What kinds of engagements are available?
- Architecture reviews, retrieval and agent system work, and production ML delivery. Engagements are selective and sized around a specific problem rather than open-ended retainers.
- Where are you based?
- I am based in Bhopal, India, and have worked remotely with teams in India and internationally.
- How do we start?
- Send a short brief describing the problem, your timeline, and the context. If it looks like a fit, the next step is a scoping conversation.
Have a problem worth describing?
A short brief is enough to start. Describe the system, what is going wrong, and the timeline you are working to.