Every capability employers ask for — and who pays for it
68 capabilities named in job postings across 17 roles, and the few you need on the way. The number on each row is the strongest ask: the share of one role's postings that name it. Not a list of everything that exists — a list of what is being hired for.
Programming foundations
6 capabilitiesWrite production-quality Python for AI work
Write clean, tested Python with virtualenvs, typing, packaging and async basics — the lingua franca of every AI role.
Build and consume REST APIs
Design and ship HTTP APIs (FastAPI/Flask/Express), call third-party APIs, handle auth, errors and rate limits.
Design scalable AI-backed systems
Reason about latency, queues, caching, storage and failure modes for systems that call models.
Ship faster with AI coding assistants
Use Copilot/Cursor/Claude Code effectively: specify, review, test and not blindly trust generated code.
Collaborate with Git and GitHub
Branch, commit, review PRs, write READMEs and keep a public portfolio employers can inspect.
Solve coding interview problems
Arrays, hashing, trees, graphs, DP at the level asked in Indian product-company screens.
LLM application development
6 capabilitiesDesign and version prompts systematically
Write system prompts, few-shot examples, output constraints; manage prompt versions and regressions.
Build an end-to-end chat assistant
A working product: UI + backend + LLM + memory + guardrails, deployed for others to use.
Integrate LLM APIs into an application
Call OpenAI/Anthropic/Gemini/Bedrock APIs with streaming, retries, token accounting and cost awareness.
Build voice or vision LLM features
Speech-to-text, TTS, image understanding, document OCR pipelines with multimodal models.
Manage context windows and memory
Token budgeting, summarization, conversation memory, caching prompts across turns.
Get reliable structured outputs from LLMs
JSON schemas, function schemas, validation with Pydantic/Zod, handling malformed outputs.
Retrieval & knowledge systems
4 capabilitiesBuild a grounded RAG application with citations
Retrieve → rerank → generate grounded answers with citations over a private corpus.
Generate embeddings and run vector search
Choose embedding models, store vectors (pgvector/Chroma/Pinecone), run similarity and hybrid search.
Evaluate and improve retrieval quality
Build eval sets, measure recall/faithfulness (RAGAS-style), tune chunking, reranking, hybrid search.
Ingest and chunk documents
Parse PDFs/HTML/Docs, clean text, choose chunking strategies, preserve metadata.
Agents & workflows
4 capabilitiesBuild a multi-step agent workflow
Plan-act-observe loops, state, retries, human-in-the-loop, with LangGraph/OpenAI Agents/Claude SDK or plain code.
Implement tool / function calling
Define tools, let the model call them, execute safely, return results into the loop.
Expose and consume tools via MCP
Build an MCP server for a data source and connect it to agent clients.
Evaluate and harden agents
Trajectory evals, cost/latency budgets, failure taxonomies, sandboxing and permissioning.
Evaluation, safety & observability
4 capabilitiesApply guardrails, safety and privacy controls
Input/output filtering, PII redaction, prompt-injection defenses, responsible-AI policies (DPDP-aware).
Build an LLM evaluation harness
Golden sets, LLM-as-judge, regression suites, scoring in CI before shipping prompt/model changes.
Trace, monitor and debug LLM apps in production
Langfuse/LangSmith/OpenTelemetry traces, token & cost dashboards, drift and failure alerts.
Optimize inference cost and latency
Model routing, caching, batching, smaller models, quantization/vLLM where self-hosting.
Cloud, deployment & production
7 capabilitiesSecure keys, auth and data in AI apps
Secrets management, IAM least privilege, OAuth, tenant isolation, audit logs.
Troubleshoot and support AI systems in production
Own L1-L3 tickets for an AI product or platform: read logs and traces, reproduce model and API failures, follow the runbook, escalate with a clean root-cause note and close within SLA.
Deploy an AI service to the cloud
Ship an LLM/ML API on Cloud Run / AWS Lambda-ECS / Azure with HTTPS, env config and logs.
Automate builds, tests and deploys with CI/CD
GitHub Actions pipelines that test, build images and deploy on merge.
Containerize an application with Docker
Write Dockerfiles, compose services, manage env/secrets, publish images.
Run workloads on Kubernetes
Deployments, services, scaling, and GPU scheduling basics for model serving.
Use managed AI platforms (Bedrock / Vertex / Azure AI Foundry)
Provision models, manage IAM, use platform features (knowledge bases, guardrails, agents) on one hyperscaler.
Machine learning & data science
8 capabilitiesBuild image models (detection/classification)
OpenCV + pretrained CNN/YOLO fine-tuning for a practical vision task.
Train and evaluate classical ML models
Scikit-learn workflow: features, train/test split, cross-validation, metrics, overfitting.
Build and train neural networks in PyTorch
Tensors, autograd, training loops, transfer learning, GPU usage.
Operate an ML pipeline (train → register → serve → monitor)
MLflow/SageMaker/Vertex pipelines, model registry, batch/online serving, drift monitoring.
Fine-tune an open LLM (LoRA/QLoRA)
Prepare instruction data, run PEFT fine-tuning on a small model, evaluate against the base.
Forecast time series
Seasonality, ARIMA/Prophet/gradient boosting, backtesting for demand/finance use cases.
Build a recommendation or ranking model
Collaborative filtering, embeddings-based retrieval, offline ranking metrics.
Solve NLP tasks (classification, NER, similarity)
Tokenization, transformers via Hugging Face, fine-tune small encoders for classification/NER.
Data engineering & analytics
7 capabilitiesBuild dashboards that answer business questions
Power BI/Tableau/Looker Studio: model data, KPIs, drill-downs, publish and present.
Turn analysis into a decision-ready story
Frame the question, pick metrics, present insights and recommendations to stakeholders.
Query and model data with SQL
Joins, aggregations, window functions, CTEs; answer business questions from relational data.
Clean and transform data with Pandas
Load messy CSV/Excel/JSON, handle nulls, joins, reshaping, and reproducible notebooks.
Build scheduled data pipelines
Airflow/dbt/Spark basics: extract, transform, load into a warehouse with tests and lineage.
Design and read A/B tests
Hypotheses, sample size, significance, guardrail metrics, reporting results honestly.
Use LLMs to accelerate analysis (text, SQL, summaries)
Text-to-SQL, classify free text at scale, summarize reviews/tickets with LLM APIs — with validation.
Creative production & media
2 capabilitiesProduce video and images with generative AI tools
Turn a brief or script into finished visuals: shot-level prompts for image and video models (Midjourney, Stable Diffusion, Runway, Kling, Veo / Google Flow, Higgsfield), voice and avatars (ElevenLabs, HeyGen), and keep characters, style and brand consistent across scenes with a prompt library you maintain.
Edit and finish AI-generated video for social formats
Assemble generated clips, voiceover, music and captions into publish-ready Reels, Shorts and ad creatives in CapCut, Premiere Pro or DaVinci Resolve: hooks in the first three seconds, pacing, aspect ratios, subtitles, and exports that pass a brand check.
Product, business & communication
10 capabilitiesApply responsible-AI and data-protection basics
Bias, transparency, consent, DPDP Act/GDPR basics, model documentation.
Discover and scope AI product opportunities
Identify where models add value, assess feasibility/data readiness, write an AI PRD with success metrics.
Run a customer discovery → POC → pilot loop
Scope a POC with a client, build/adapt it fast, demo it, handle objections, hand over to production.
Explain how LLMs work and where they fail
Tokens, context, hallucination, RAG vs fine-tuning, cost drivers — well enough to make decisions with engineers.
Design and deliver AI training sessions
Turn a GenAI or ML topic into a session plan — slides, a hands-on lab and an assessment — and deliver it to a classroom or corporate cohort, tracking completion in an LMS.
Translate business requirements into an AI solution design
Discovery, BRD/FRD, process mapping, solution architecture diagrams and success criteria.
Communicate AI trade-offs to stakeholders
Write and present clearly about risk, cost, timelines and limitations; run demos and reviews.
Define quality metrics and eval plans for AI features
Choose offline/online metrics, design human-rating rubrics, set launch bars for AI features.
Deliver technical demos and answer RFPs
Build demo environments, tailor narratives per persona, write technical responses to RFP/RFI.
Prototype an AI feature without an engineering team
Build a clickable/working prototype with no-code tools, prompt playgrounds or AI coding assistants.
AI automation & no-code
5 capabilitiesAutomate workflows with n8n / Zapier / Make
Triggers, actions, branching, error handling, connecting SaaS tools via APIs and webhooks.
Add LLM steps to business automations
Classify emails, extract fields from documents, draft replies inside n8n/Zapier/Power Automate flows.
Map a process and quantify automation ROI
Document current process, find automatable steps, estimate hours saved, propose the workflow.
Automate legacy UI tasks with RPA (UiPath / Power Automate Desktop)
Record/replay bots, selectors, exception handling for desktop and browser tasks.
Deploy a support/sales chatbot on WhatsApp or web
Voiceflow/Botpress/custom: intents, knowledge base, handoff to human, analytics.
Annotation, quality & human feedback
7 capabilitiesLabel data accurately against guidelines
Read a labeling spec, apply it consistently (text/image/audio), flag ambiguous cases.
Audit label quality and compute agreement
Sampling, inter-annotator agreement, error taxonomies, feedback to annotators.
Annotate and evaluate in an Indian language
Be the native-language expert on a dataset: label, transcribe, rate and culturally adapt content in Hindi, Tamil, Bengali, Malayalam or another Indian language, and judge whether a model's output is fluent, accurate and culturally appropriate for that audience.
Evaluate AI outputs as a domain expert
Use professional expertise (healthcare, BFSI, law, software, geospatial) to judge whether a model's answer is correct in that field, and write a rationale a non-expert reviewer can follow. This is what separates low-paid labelling from expert evaluation work.
Evaluate and rank model responses (RLHF / preference data)
Compare responses for helpfulness, accuracy, safety; write rationales; follow rubrics.
Write labeling guidelines and evaluation rubrics
Turn a vague quality goal into a rubric with examples and edge cases that others can apply.
Work in annotation tools (Label Studio / CVAT / Labelbox)
Set up projects, hotkeys, bounding boxes/segmentation/NER, export formats.
Career signal & proof
2 capabilitiesPublish a portfolio employers can verify
2–3 deployed projects with READMEs, demo links and architecture notes on GitHub; LinkedIn that mirrors it.
Clear the AI-role interview loop
Explain past projects, answer role-specific technical questions, take-home tasks and system/product design rounds.
Search & AI visibility
3 capabilitiesGet a brand cited in AI answers (GEO / AEO)
Find the prompts buyers ask ChatGPT, Gemini, Perplexity and Google's AI Overviews, track whether and how the brand is cited, then reshape content, sources and crawler access so it is the answer — and report share of voice over time.
Run technical and on-page SEO for a website
Audit a site the way search engines and AI crawlers read it: crawlability, indexing, Core Web Vitals, internal links and on-page signals, using Search Console and an SEO suite, then fix what blocks it from being found.
Implement schema and entity markup
Describe a brand, its products and its answers in schema.org JSON-LD so search engines and AI assistants can identify the entity and quote it correctly; validate it and keep it in step with the page.
Shares are per role: the percentage of that role's job postings in India that name any skill mapped to the capability. Counting rules are on the methodology page; the full table is open data. Prefer to start from a job? All 17 roles →