Ai Engineer in New Delhi, भारत

About Drishti IAS

Drishti IAS is one of India’s most trusted institutions for UPSC Civil Services preparation, serving lakhs of aspirants through classroom programmes, a bilingual (Hindi and English) digital platform, daily current-affairs content, test series and video learning. We are bringing intelligent, reliable and multilingual learning experiences to our students and making our editorial, operations and support workflows dramatically more efficient through AI.

About the Role

We are looking for a hands-on AI Engineer with 1–3 years of experience who can take LLM-based systems from idea to production. You will split your time roughly equally between two tracks: student-facing AI products such as doubt-resolution assistants, answer-writing evaluation, personalised study aids and voice/vision-enabled learning; and internal automation that accelerates content creation, data pipelines and business operations. You will own the full stack of an AI feature: data, retrieval, models, serving, evaluation and monitoring.

Key Responsibilities

• LLM & RAG systems: Design, build and maintain retrieval-augmented pipelines over Drishti’s large content corpus, including notes, current affairs, PYQs and test series. Own chunking, embedding, hybrid search, re-ranking and citation quality, and set up offline/online evaluation and production monitoring for accuracy, hallucination and latency.

• Agentic workflows & automation: Build tool-using agents with LangChain/LangGraph or similar SDKs, expose capabilities via MCP, and automate editorial, marketing and support workflows on platforms such as n8n, Dify or Zapier.

• Inference & deployment: Deploy and serve open-weight models on GPU infrastructure using vLLM, Transformers or llama.cpp. Handle quantisation, batching, KV-cache and throughput, latency and cost trade-offs.

• Fine-tuning: Create high-quality datasets from in-house material and fine-tune LLMs using SFT, LoRA, QLoRA and preference tuning for domain, tone and Hindi-English tasks. Evaluate rigorously against baselines before release.

• Multimodal integration: Integrate speech tools for STT/TTS, including Whisper and ElevenLabs, for voice-based learning and dictation. Integrate image-generation and editing models, including diffusion models and Nano Banana, for content and creative workflows.

• Data engineering: Build pipelines that ingest and process large volumes of text, audio and long documents. Design schemas and queries across PostgreSQL/MySQL and vector databases such as pgvector, Qdrant and Milvus.

• System design & reliability: Design scalable, observable services with clean APIs. Add logging, tracing, cost tracking and guardrails. Apply secure-by-default practices, including prompt-injection defences, secrets management and least-privilege access for tools and agents.

• Collaboration: Work closely with subject-matter experts, editors and product stakeholders, and document systems clearly.

Required Skills & Experience

• Python: Strong, production-quality Python skills, including typing, async, packaging and testing. Comfortable with FastAPI or a similar framework.

• Automation platforms: Hands-on experience with n8n, Dify, Zapier or comparable low-code/workflow tools, including custom nodes and webhooks.

• RAG end-to-end: Experience building, evaluating and monitoring RAG systems using RAGAS, custom evaluations or LLM-as-a-judge approaches. Strong SQL skills with PostgreSQL/MySQL and practical experience with vector databases.

• Agent frameworks: Experience with LangChain, LangGraph or similar SDKs/ADKs, such as Google ADK, OpenAI Agents SDK or Pydantic AI. Knowledge of function/tool calling and Model Context Protocol (MCP).

• System design: Ability to design and reason about services, queues, caching, data flow and failure modes for AI applications.

• Inference engineering: Experience deploying local/open-weight models using vLLM, Transformers or llama.cpp. Knowledge of GPUs, VRAM budgeting, quantisation formats such as GGUF, AWQ and GPTQ, and CUDA basics is a plus.

• Fine-tuning: End-to-end experience in dataset construction and cleaning, training using PEFT, TRL, Unsloth or Axolotl, and evaluation.

• Vision & voice models: Experience integrating diffusion/image models, such as Stable Diffusion, FLUX and Nano Banana, along with STT/TTS models, such as Whisper, ElevenLabs and Indic TTS, into applications.

• Large-scale & long-context data: Experience handling high-volume corpora and long sequences through batching, streaming, context management and long-context model usage.

• Agentic systems: Experience with multi-step agents, tool usage, memory, planning loops and safe execution boundaries.

• ML/DL fundamentals: Basic understanding of machine learning and deep learning, including training loops, loss functions, embeddings, transformers and evaluation metrics.

Good to Have

• Experience with Hindi or other Indic-language NLP, transliteration and multilingual embeddings.

• Experience with Docker, Linux, CI/CD, Git workflows and cloud GPU providers such as AWS, GCP, Azure or RunPod.

• Familiarity with LLM application observability tools, such as Langfuse, LangSmith and Arize Phoenix, and prompt/version management.

• Awareness of AI security, including OWASP Top 10 for LLM Applications, prompt-injection and data-exfiltration risks, output filtering and red-teaming.

• Open-source contributions, personal projects, blogs, Kaggle work or hackathon experience in GenAI.

Who You Are

• You are ship-oriented and prefer a working prototype today over a perfect design next month, followed by iteration using evaluations.

• You are curious and current on model releases, benchmarks and tooling, and understand when not to use an LLM.

• You are rigorous about quality, measure before making claims, and treat hallucinations, latency and cost as bugs.

• You are a clear communicator who can explain technical trade-offs to non-technical educators and stakeholders.

What We Offer

• Ownership of AI products used by lakhs of UPSC aspirants, with real-world impact on learning outcomes.

• Access to dedicated GPU infrastructure, a rich proprietary content corpus and freedom to experiment with open-weight models.

• Competitive compensation, a learning budget and a growth path towards Senior AI Engineer or AI Lead roles.