Hire an AI Engineer in India
Hire an AI engineer in India who ships production AI: multi-LLM orchestration, RAG, pgvector semantic search, and voice/SMS agents on reliable backends. Remote, IST.
Looking to hire an AI engineer in India who ships production systems, not demos? I’m Kuldeep Pisda — a senior backend and AI engineer and former startup CTO based in Bengaluru. For 6+ years I’ve put AI in front of real users: multi-LLM orchestration behind a unified layer, retrieval over hundred-thousand-document knowledge bases, semantic search on pgvector, and agentic voice and SMS automation — all on backends built to stay up, stay cheap, and stay correct. The interesting part of AI work isn’t the prompt; it’s everything around it.
What I build
- LLM agents — task-oriented agents with tool use, structured outputs, and guardrails, wired into real business workflows rather than a chat box.
- RAG and semantic search — retrieval pipelines over large document sets, chunking and embedding strategies, and re-ranking that actually improves answers. I’ve tuned pgvector for 100k+ documents and built Indic/Hindi semantic search with multilingual SBERT.
- Voice and SMS agents — agentic phone and messaging automation with Deepgram speech-to-text, turn-taking, and fallbacks for when a model stalls mid-call.
- Multi-LLM orchestration — Claude, GPT-4o, Gemini, and Perplexity behind one unified layer, with retry/backoff, provider failover, and per-tenant model selection so each customer runs the model that fits their cost and quality bar.
- Evaluation and reliability — prompt pipelines with versioning, eval sets to catch regressions, latency and cost budgets, and observability so you know why a response was wrong, not just that it was.
Experience you’re hiring
I’ve built AI features that carry real load, described here by role since the clients are under wraps — the case studies walk through the decisions and trade-offs:
- A multi-LLM due-diligence engine that fans a single query across several model providers behind one layer, with failover and per-tenant model choice, to score and match at scale.
- A RAG assistant over 100k+ documents — the retrieval, chunking, and re-ranking work that makes answers grounded and citable instead of confidently wrong.
- Hindi semantic search on pgvector, tuned for an Indic language where naive embeddings fall apart, serving a platform I scaled past 30,000 users.
The backends under all of this are the boring, load-bearing part: Django/DRF and FastAPI, PostgreSQL with pgvector, Celery for background work, and event-driven orchestration with Inngest. I wrote django-rls for multi-tenant isolation, and I speak at DjangoCon US, DjangoCon Europe, and EuroPython. I test what I ship.
How engagements work
Short, well-scoped engagements with a written plan and testable milestones — a fixed-scope AI build, an ongoing retainer, or an architecture review before you commit to a direction. Remote-first and async-friendly, based in India (IST) and used to overlapping with US and EU hours.
FAQ
What does “production AI” actually mean? It means the feature survives real traffic: it handles a model timing out, a provider rate-limiting you, a malformed response, and a cost spike — without paging you at 2am. Demos ignore all of that. Production is the difference.
Which models do you work with? Claude, GPT-4o, Gemini, and Perplexity, plus Deepgram for speech-to-text. I put them behind a unified layer so switching or mixing providers is a config change, not a rewrite — and so each tenant can run the model that fits their needs.
RAG or fine-tuning? Usually RAG. For most business problems, grounding a strong general model in your own documents is cheaper, faster to update, and easier to debug than fine-tuning. I’ll tell you honestly when fine-tuning genuinely earns its cost — it’s rarer than the hype suggests.
How do you handle cost and latency? By treating them as budgets, not afterthoughts: caching, right-sizing the model per task, streaming responses, and measuring token spend per request so it doesn’t quietly balloon. Cheaper and faster usually come from architecture, not from a smaller model.
Can you add AI to an existing product? Yes — most of my AI work bolts onto live systems. I’ll map your data and traffic first and integrate carefully, rather than dropping in an unbounded LLM call and hoping.
Let’s talk
Tell me what you’re trying to build and where it’s stuck — get in touch, or grab a time below:
Related: Hire a Python Developer in India · Hire a Backend Engineer in India · Hire a Fractional CTO in India · Fractional CTO for Startups · Hire an LLM Engineer