Prompt Engineer / AI Workflow Specialist
Also posted as: Prompt Engineer · AI Prompt Specialist · AI Content Engineer · Conversation Designer · AI Specialist
A Prompt Engineer in India writes and versions the instructions that make an LLM behave: system prompts and few-shot examples for a document-extraction or support workflow, strict JSON schemas so downstream code can parse the output, a golden set of test cases, and a weekly loop of reading failures and fixing prompts. The hiring is led by IT-services and GCC firms (Accenture, Infosys, TCS, PwC AC India) plus conversational-AI and voice-bot startups in Bengaluru, NCR and Hyderabad, and the pay splits in two rather than sitting on a ladder — 4-8 LPA for the voice, content and enablement postings, 30-50 LPA for the rare engineering ones. Be clear-eyed about what you are aiming at: the title is thinning (Glassdoor India's own count fell from 5,995 to 5,229 in a month), 13 of these 32 postings carry no 'prompt' in the title, and the work is dissolving in both directions — down into conversation design and enablement, up into agent engineering — so treat prompting as your entry wedge and add evals, structured outputs, Python and agent workflows fast: that is the half of the split that pays.
- Across 113 Prompt Engineer / AI Workflow Specialist job postings in India, the most-requested capabilities are Design and version prompts systematically (85%), Build an end-to-end chat assistant (73%) and Write production-quality Python for AI work (66%).
- Pay at 0–2 yrs averages about ₹6 LPA (Glassdoor only, not cross-checked yet); most earn ₹4–8.5 LPA (base pay); employers offer ₹3.5–5.2 LPA (median of 12 job postings that state pay, 0–2 yrs · other sites, LinkedIn, Cutshort, Wellfound).
- At least 107 open roles in India — the dated job postings for this role we read in the last 60 days; portals count a title's exact phrase, which undercounts a job posted under many titles, checked 03-10-2026.
- Hiring is concentrated in Bengaluru, Delhi NCR and Hyderabad.
- Postings read from LinkedIn 81%, other portals 16%, cutshort 1%, Wellfound 1% and company career pages 1% — one portal supplies most of this sample, so the shares lean to the employers that post there.
Hire for this role? Add your read to this page — what decides the offer, what it closes at. An email to Ajeet, ten minutes, credited or not as you choose. How that read is shown.
Capabilities employers ask for
How often employers ask for each capability, measured across the job descriptions behind this page. Click one to see what "knowing it" means, how to learn it, and how to prove it.
Have a posting open? Check it against this map →
Design and version prompts systematically
This is the job's core, and postings say so plainly: Accenture lists prompt engineering as the must-have skill, Rex.zone wants product requirements translated 'into system prompts, developer prompts, and user patterns', and mea asks for few-shot and zero-shot technique by name. Bain and AXA XL push it further, wanting reusable prompt frameworks the rest of the company can pick up. You need to write, test and version prompts as a library with templates and change notes, not as one-off chat messages.
Explore 4 practice tools →Build an end-to-end chat assistant
A lot of this work is making an assistant behave in front of real users. Foundever wants conversational flows 'for chatbots and voicebots', PwC wants intents, entities and dialog flows on contact-centre platforms, JoyzAI turns client requirements into 'prompt strategies, conversation flows, and guardrails' for a calling agent and chatbot, and BMC wants AI chatbots and virtual assistants deployed. Be able to ship a small assistant with a persona, fallback answers and a hand-off to a human, and demo it.
Explore 6 practice tools →Write production-quality Python for AI work
Python shows up across the board, but the bar swings widely: mea asks for 'basic Python knowledge (for testing / scripting)', while HARTING wants 'production Python code integrating LLMs and vector stores' and RingCentral marks strong Python as mandatory. SS&C sits in the middle, wanting Python 'to programmatically generate, test, and evaluate prompt performance at scale'. Learn enough to script a batch of prompt tests against an API and read the results; the engineering-flavoured postings expect more.
Explore 4 practice tools →Integrate LLM APIs into an application
Employers assume you work with models as components rather than a chat window, though how deep that goes varies. HARTING wants 'practical experience integrating LLMs', PwC wants agents capable of 'tool/API invocation', and Worldwide Clinical Trials wants multi-step prompts that run across Claude and Microsoft Copilot in real workflows. At minimum, call at least one provider's API from code, set parameters like temperature deliberately, and know what a request costs.
Explore 4 practice tools →Build a multi-step agent workflow
Prompting increasingly happens inside agents. HARTING wants you to 'architect and implement multi-agent orchestration', Bain wants agents that 'orchestrate multi-step workflows, tool calling', EY wants Copilot Studio agents with actions and orchestration logic, and project44 wants agent workflows for carrier monitoring and escalation. Learn the loop of plan, call a tool, read the result and hand off, even if your title still says prompt.
Explore 4 practice tools →Communicate AI trade-offs to stakeholders
Prompts are writing, and employers screen for it: Synechron and Navi both ask for strong written and verbal communication, XpertDox wants 'clear written communication across time zones', and KeyValue wants writing clear enough 'to structure precise instructions'. Quantiphi's conversational-AI analyst role adds managing customer communication outright. You must be able to explain what a prompt does, why it failed and what you changed, in plain English to people who will never read it.
Explore 3 practice tools →Build a grounded RAG application with citations
Grounding comes up often, usually as something you work with rather than build from scratch. Rex.zone wants you to 'reduce hallucinations using RAG strategies and citation constraints', Cognizant wants RAG pipelines and vector stores 'so agents are grounded', and RingCentral wants RAG over enterprise knowledge sources. Know how retrieval feeds a prompt and how to make answers cite their source; index internals can stay light.
Explore 5 practice tools →Build an LLM evaluation harness
Employers want proof that a prompt got better, not a feeling that it did. Fello asks for evaluation frameworks with 'regression test suites from real production data', Capco for prompt benchmarking and regression testing, and Sarvam wants you to 'define what good means for the use case, build the evals that measure it, and close the gap'. Be able to build a small golden set, score every prompt change against it, and show the before-and-after.
Explore 4 practice tools →Apply guardrails, safety and privacy controls
Safety lives in the prompt layer here, and employers name it: PwC applies 'prompt engineering, guardrails, and evaluation frameworks' so answers stay 'accurate, safe, and on-brand', SS&C wants guardrail frameworks such as NeMo Guardrails enforcing safety rules, and JoyzAI has you read call transcripts daily to catch hallucinations. You should be able to spot failure cases, add refusals and checks that block them, and prove with tests that they hold.
Explore 3 practice tools →Add LLM steps to business automations
Some postings put the model inside a business process rather than a product. Bain wants Python integrations so LLMs handle 'summarization, classification, extraction, reasoning, and workflow automation', Persistent Systems wants Power Automate flows behind Copilot actions, and hodos360.ai wants n8n workflow automations with AI steps. Be able to drop a model call into an automation tool, route its output to the next step, and handle the case where it answers badly.
Explore 5 practice tools →Build voice or vision LLM features
Not all of this work is text. PwC wants IVR call flows tuned down to 'DTMF/ASR grammar, barge-in', Foundever asks for speech technologies (STT/TTS), and Huvo AI configures voice agents and WhatsApp workflows, while QuantumQuake, Mushroom World and WPP want prompting for image and video tools. Pick one lane, voice or visual, and be able to show a working prompt set built for it.
Explore 4 practice tools →Manage context windows and memory
A smaller group of postings names context, tokens and cost directly. Takeda's Context Trainer runs 'A/B experiments across vector stores' while watching answer quality against cost and latency, Aceolution wants you to 'optimize model context, token usage, response latency', and Sarvam lists context engineering and memory among its core asks. Know how context windows fill up, what to trim or summarise, and how to measure the cost of each choice.
Explore 4 practice tools →Get reliable structured outputs from LLMs
The engineering-track postings need answers another program can read. mea asks for 'JSON, structured outputs, schema-driven extraction', Data Eminence for 'structured outputs and JSON-based AI responses', and Sarvam and Chryselys list structured outputs next to tool use and memory. The content-track postings never ask, which is exactly why it separates the two: be able to make a model return valid JSON against a schema and recover cleanly when it doesn't.
Explore 4 practice tools →Write labeling guidelines and evaluation rubrics
Very few postings ask for the rubric itself, but the ones that do are specific: Rex.zone wants evaluation plans 'using rubrics for accuracy, completeness, instruction following, tone, and policy compliance', and Fello asks for 'evaluator prompts, rubrics' alongside datasets. The nearest everyday ask is Capco's 'define evaluation criteria and acceptance testing for AI-generated outputs'. Be able to write a scoring guide another person, or a model, can apply consistently.
Explore 3 practice tools →Families: LLM application development · Programming foundations · Agents & workflows · Product, business & communication · Retrieval & knowledge systems · Evaluation, safety & observability · AI automation & no-code · Annotation, quality & human feedback
"AWS" on a JD is not "learn AWS"
The words employers write, translated into what they want you to be able to do for this role.
Skip, for now
- Fine-tuning / LoRA / RLHF training — Named in 6 of 32 job postings (Synechron, TCS 'preferred', Rex.zone, HARTING's tool list, mea's RLAIF feedback loop, Mushroom World) — and every one of them wants you to feed a training process, not run one: Rex.zone asks you to create preference data, Mushroom World to give feedback for fine-tuning. Prompting plus a good eval set beats fine-tuning for almost every job on this list.
- Deep learning and transformer internals (PyTorch, training loops) — Deep learning is named in exactly 1 of 32 job postings (Synechron, at 5+ years), machine learning or NLP in 5. Know tokens, context windows, temperature and why hallucinations happen — that is the depth these 32 job postings actually test.
- Docker, Kubernetes and cloud deployment — Appears in 4 of 32 job postings and is genuinely required in one: HARTING wants AWS, Docker and CI/CD, which is an engineering req wearing a prompt title; Huvo AI lists it as nice-to-have, PwC and Takeda in passing. Ship your project on a free tier and move on — unless you are chasing the 30-50 LPA engineering postings, infra is not what gets a prompt engineer hired in India.
- Vector database internals and embedding-model benchmarking — RAG shows up in 9 of 32 job postings, but in seven the wording is 'collaborate with AI Engineers on RAG' (LLM Decode) or 'knowledge of RAG' (Foundever); only HARTING and Takeda ask you to build or tune the pipeline yourself. Learn to use one vector store and measure retrieval quality; skip index tuning and self-hosted embedding servers.
- Paid prompt-engineering certifications and prompt-pack marketplaces — Zero of 32 job postings ask for a certificate. LLM Decode asks for a 'portfolio of prompt engineering projects', Streevia for GenAI projects or coursework, QuantumQuake for a portfolio of AI-generated work — build artefacts with before/after numbers instead.
Prompt workbench: a document-extraction assistant with a versioned prompt library and an eval harness
Pick a messy real-world document set (insurance loss runs, invoices, college transcripts, WhatsApp support transcripts) and build a small Python service that extracts a fixed schema from each document using an LLM. Keep every prompt in a versioned file with a changelog, force output through a Pydantic schema with a retry-and-repair path, and build a golden set of 60 hand-labelled cases scored for field accuracy plus an LLM-as-judge rubric for tone and refusals. Then run the same suite across two models (say GPT and Claude, or one open-weight model), publish a comparison table with cost and latency per document, and put a simple chat or voice front end on it so a non-technical user can try it. This mirrors mea, Prasanz and Rex.zone almost line for line.
- Prompt library is versioned in git with a README showing at least three iterations and the accuracy each version scored
- Every output validates against a JSON schema; malformed outputs are caught, repaired or logged rather than crashing the pipeline
- Eval harness runs from one command over 60 labelled cases, reports per-field accuracy plus a rubric score, and flags regressions when a prompt or model changes
- A model-comparison table reports quality, cost per 1,000 documents and p50 latency, with one paragraph on which you would ship and why
- Guardrail path is tested: at least five adversarial or out-of-scope inputs (prompt injection, missing fields, wrong language) produce a safe refusal instead of an invented answer
What the interviews look like
The rounds you'll actually face, in the order they usually come.
- 1
Screening
Recruiter or hiring manager checks written English first (this role is judged on writing), which models and platforms you have used hands-on, whether you can code enough Python for testing, and salary expectations — be ready for 4-8 LPA at entry, and know that the 30-50 LPA postings (XpertDox, HARTING) are engineering interviews wearing a prompt title. Have a portfolio link with prompts and eval numbers, not screenshots of chats.
- 2
Technical / take-home
A failing prompt or a raw document set is handed to you: improve accuracy, force JSON output against a schema, and explain each change. Expect questions on few-shot vs zero-shot, chain-of-thought, temperature and top_p, context windows and token cost, and how you would detect a regression when the vendor updates the model.
- 3
System / product design
Design the loop around the prompt: golden set and rubric, LLM-as-judge vs human review, versioning and rollback, guardrails against injection and PII leakage, human-in-the-loop corrections, and where the prompt sits inside a RAG or multi-agent flow. Conversational-AI employers will instead ask you to design a full dialog flow with fallbacks, handoff and containment metrics.
- 4
Culture / stakeholder
Product, delivery or client leads probe how you explain LLM limits to a non-technical stakeholder, how you handle a hallucination reported by a customer, and domain comfort (BFSI, insurance, contact centre, ed-tech). At IT-services firms and voice-bot startups expect a question on Indian-language coverage — Hindi plus one regional language is a real advantage.
What people ask before choosing this role
Can a fresher get a Prompt Engineer / AI Workflow Specialist job in India?
Yes, this is one of the more reachable AI-era roles. 36 of the 113 job postings behind this page accept 0–2 years of experience. The rest want more, so expect the fresher-friendly openings to be competitive.
What is the salary of a Prompt Engineer / AI Workflow Specialist in India?
Pay at 0–2 yrs averages about ₹6 LPA (Glassdoor only, not cross-checked yet); most earn ₹4–8.5 LPA (base pay); employers offer ₹3.5–5.2 LPA (median of 12 job postings that state pay, 0–2 yrs · other sites, LinkedIn, Cutshort, Wellfound). Not every posting states pay, and pay varies widely by city and by whether the employer is an IT-services firm, a global capability centre or a product startup.
How long does it take to become a Prompt Engineer / AI Workflow Specialist?
The six capabilities employers ask for most add up to roughly 100 focused hours — about 13 weeks at 8 hours a week, if you are starting from zero on all of them. Most people are not: the self-check on this page works out what you can skip, which is usually a large part of it.
What skills do you need for a Prompt Engineer / AI Workflow Specialist role?
Across the 113 job postings behind this page, the most-requested capabilities are Design and version prompts systematically (85% of postings), Build an end-to-end chat assistant (73% of postings) and Write production-quality Python for AI work (66% of postings). Note these are capabilities, not tools — employers write tool names, but what they are buying is the ability to do the work.
Which cities in India post the most Prompt Engineer jobs?
Bengaluru (34), Delhi NCR (19), Hyderabad (14) and Remote (India) (13) — counted across the 113 job postings behind this page. Remote-India roles are counted separately where the posting said so.
Is demand for Prompt Engineer roles in India growing?
The skill is growing while the standalone title shrinks: Recrew.ai reports job postings titled 'prompt engineer' down about 40% from the 2023 peak, AmbitionBox shows exact-title pay flat to down (typical ₹5.9-6.6 L/yr from 355 salaries, with a 4% drop in average salary over two years), and 36 of the 64 postings we collected since 1 September 2026 carry no 'prompt' in the title at all (Copilot Engineer, Agent Engineer, AI Context Engineer, Conversational AI Developer). Overall AI/ML hiring rose 31% year on year in August 2026 (Naukri JobSpeak), so the work is in demand; it mostly arrives under another title, and the postings that fold prompting into conversational-AI or agent engineering list ₹10-50 LPA.
Do I need a degree or a paid certificate for this?
Nothing on this page requires a paid certificate, and none of the 113 job postings behind it asked for one by name. What they ask for is evidence you can do the work — a public repo, a deployed project, something a hiring manager can open. That is what the path on this page is built to produce.
Companies with this role open in India
A sample of employers we saw hiring for this role — IT services, global capability centres, product companies and startups. Each links to one of the company's postings for this role, checked open on 04-10-2026; where none is open, to its current openings instead.
What it pays
most earn ₹4–8.5 LPA (base pay) · Glassdoor only, not cross-checked yet
Employers offer ₹3.5–5.2 LPA: median of 12 job postings that state pay, 0–2 yrs · other sites, LinkedIn, Cutshort, Wellfound
Not enough salary data yet for Mid · 2–5 yrs or Senior · 5+ yrs.
Across all levels: the middle half earns ₹4.2–9.1 LPA · Glassdoor
How much demand
What each job portal shows for this role's title — the readings behind the openings figure above.
- at least 3131 of the first 580 results carry the title (reading stopped: error: empty page at start=610 although headline total is 9,000+) · checked 04-10-2026linkedin
- at least 2424 of the first 43 results carry the title (reading stopped: error: sign-in wall after page 1) · checked 04-10-2026indeed
- at least 9696 of the first 982 results carry the title (reading stopped: cap) · checked 04-10-2026naukri
- 107dated job postings for this role we read in the last 60 days — the figure above, because no portal's count reached it
Where this role is heading
- The skill is growing while the standalone title shrinks: Recrew.ai reports job postings titled 'prompt engineer' down about 40% from the 2023 peak, AmbitionBox shows exact-title pay flat to down (typical ₹5.9-6.6 L/yr from 355 salaries, with a 4% drop in average salary over two years), and 36 of the 64 postings we collected since 1 September 2026 carry no 'prompt' in the title at all (Copilot Engineer, Agent Engineer, AI Context Engineer, Conversational AI Developer).
- Overall AI/ML hiring rose 31% year on year in August 2026 (Naukri JobSpeak), so the work is in demand; it mostly arrives under another title, and the postings that fold prompting into conversational-AI or agent engineering list ₹10-50 LPA.
Capability percentages come from 113 job descriptions read in full on 03-10-2026. How we do this