AI engineer in India — Deepak Kumar

At India Today Group — the publisher behind Aaj Tak — I build Generative AI into editorial products: article generation, AI summaries and semantic search over the archive, using OpenAI and LangChain with retrieval on MongoDB Atlas Vector Search. The hard part is never the model call — it is chunking tuned to the corpus, evaluation before rollout, and an editorial review step so nothing unchecked reaches a reader.

• AI podcast platform — articles to broadcast-ready audio, cutting production time roughly 80%
• RAG over an editorial archive, with retrieval quality measured rather than assumed
• Semantic search and AI summaries built into tools journalists use daily
• Structured prompting behind typed API contracts, so failures are catchable in code

Like any other dependency: with a latency budget, a cost budget, a fallback when it fails, and monitoring that names the failing component. An LLM call is a network call to a probabilistic service — treating it as anything more magical is how teams end up with features they cannot debug or afford.

• Evaluation sets built from real queries before a feature is promoted
• Guardrails on output, plus a human review step wherever output is published
• Cost and latency tracked per call, alerting like any other service dependency
• Graceful degradation — the product still works when the model is slow or down

An AI engineer who cannot build the product around the model tends to hand over a prototype. I bring 9+ years of MERN and Next.js behind the AI work, which means the retrieval pipeline, the API, the dashboard and the deployment come from the same person.