Software Engineers
Moving from building features to owning AI delivery with customers.
Forward Deployed Engineering Practitioner Program
Become the engineer who takes AI from a promising prototype to a working system inside a real organisation.
Online learning with three optional campus immersions at IIT Hyderabad.
Revised schedule: At participants’ request, no sessions will be held during the Christmas and Sankranti weeks (26 December 2026 and 16 January 2027). The programme now concludes on 14 March 2027.
See the scheduleMap the customer workflow, constraints and success measures before choosing the technology.
Build services, data flows, AI workflows and enterprise integrations in one engagement.
Release safely, operate reliably and connect user adoption to measurable business value.
Bring your engineering foundation. Build towards delivery ownership, from understanding a customer’s problem to deploying and handing over a working AI solution.
Moving from building features to owning AI delivery with customers.
Who can build models, pipelines and prototypes, and want them running in production.
Hands-on technologists stepping into customer-facing AI delivery.
Taking AI pilots to production together. [Team terms]
Write, read and debug app code; packages, exceptions, structured data.
HTTP, REST and JSON; FastAPI and Pydantic.
SQL, tables, keys and basic transactions in Postgres.
Git, Linux terminal, dependencies, existing repos.
Basic pytest; running a containerised app.
Every lab adds to one engagement repository, from customer discovery to operational handover.
Customer brief, workflow map, requirements, risks and an outcome hypothesis.
Reproducible repository, typed service, versioned API, data layer, tests and CI.
Integrations, event processing, model routing, retrieval and agent workflows with human control.
Role-aware operator interface, deployment architecture, enterprise identity and network fit.
Tracing, service-level objectives, evaluation harness, threat model and failure exercises.
Staged rollout, adoption and value measurement, handover, executive readout and postmortem.
The capstone connects the customer workflow, architecture, integrations, failure modes, evaluation, operations, adoption and business outcomes. A chat interface alone does not meet the engagement standard.
Each module adds evidence to the same engagement repository. The curriculum is a draft; the session-to-date mapping is being finalised.
24 sessions, guided labs and a capstone
Model, reasoning, context and gateway engineering. Agent harness, skills and engineering velocity.
Advanced retrieval strategies. Knowledge bases, knowledge graphs and GraphRAG.
Loop engineering. Graph engineering and adaptive agent topologies.
Memory and long-context engineering. Agent evaluation and neuro-symbolic verification.
Enterprise data discovery and agent-ready access. Enterprise context, semantics and ontology engineering.
Intelligent multimodal ingestion and data contracts. AI storage architecture and knowledge lifecycle.
Semantic-layer-grounded SQL agents. Analytical intelligence across structured and unstructured data.
Agent-ready enterprise APIs, MCP and connectors. Event-driven, proactive and recoverable agent workflows.
Observability, continuous quality and AI FinOps. Progressive autonomy, model migration and safe releases.
Enterprise identity and agent authorisation. Agentic security and AI supply chain.
Model selection, optimisation and serving. Deployable AI applications and agent-to-user interfaces.
Scaling and isolating agent workloads. Distributed inference and capacity engineering.
Over 9 weeks, you build one customer AI engagement, from problem discovery to a tested deployment, user acceptance and handover. Each week adds to the same engagement repository.
Fast vs deliberative models, structured outputs, context budgeting and compaction, caching, model routing, cost and provider abstraction.
Pi, AGENTS.md, SKILL.md, plugins, progressive disclosure, Superpowers, SDD/TDD, independent review and sandboxed execution.
Document AI, multimodal parsing, lexical, vector and hybrid search, reranking, hierarchical and vectorless retrieval, provenance, citations and abstention.
Entities, relationships, ontologies, small-scale graph construction, multi-hop reasoning, temporal facts and graph-assisted retrieval.
ReAct, plan-act-verify, bounded goal loops, heartbeat, schedule and event triggers, tool permission boundaries, budget limits, stuck detection, retry, pivot, escalation and termination.
LangGraph state, deterministic and validator nodes, conditional edges, checkpointing, human interrupts, parallel delegation, static vs adaptive graphs and A2A concepts.
Working, episodic, semantic and procedural memory, consolidation, compaction, contradiction handling, revision, expiry, deletion and poisoning defence.
Task and trajectory evaluation, calibrated judges, tool-action correctness, property testing, deterministic invariants, Z3 constraints and evaluation-driven optimisation (GEPA).
Data estates, metadata discovery, ownership, lineage, permissions, catalogues, APIs, databases, warehouse access and synthetic data environments.
Canonical entities, cross-system identity mapping, business vocabularies, semantic metrics, entity relationships, source precedence and knowledge provenance, with an introduction to Apache Ossie and metadata-aware AI.
LLM-powered extraction, classification and enrichment, deterministic transformations, schema validation, AI data contracts, exception handling, lineage, freshness and deletion propagation.
Relational, vector and graph stores, lakehouse and warehouse integration, indexing strategies, tenant-aware filtering, incremental updates, re-embedding, versioning, retention and safe migrations.
Metadata-guided schema discovery, business metric definitions, text-to-SQL, query decomposition, SQL AST validation, schema and column authorisation, read-only execution, query budgets and execution-accuracy evaluation.
Multi-step analytical planning, sandboxed Python computation, warehouse and lakehouse access, graph, SQL and vector reasoning, source attribution, numerical reconciliation and analytical evaluation.
Converting existing APIs into safe agent capabilities, OpenAPI-to-MCP adaptation, skills and plugins, typed tool contracts, scoped permissions, approval-bound actions, idempotency and connector governance, with an introduction to browser and computer-use fallback.
Webhooks, schedules, CDC, event-triggered loops, asynchronous execution, knowledge freshness, duplicate events, retries, dead-letter handling and recovery.
OpenTelemetry and OpenObserve, model, agent and tool traces, task quality, drift, cost per successful task, tenant and feature attribution, budgets, SLOs and FOCUS-aligned cost reporting.
Shadow, human-approved and scoped autonomous execution, OpenFeature and flagd, evaluation-based canaries, prompt, model and skill versioning, rollback, release manifests, technical UAT and incident recovery.
OIDC, SAML and SCIM, authentik, scoped and delegated credentials, user-to-tool attribution, MCP authorisation, enterprise-managed authorisation, RBAC and ReBAC, and gateway policy controls.
OWASP LLM, agentic and MCP risks, malicious skills and plugins, memory poisoning, tool escalation, sandbox boundaries, PII, model provenance, approved capabilities and AI bills of materials.
Hosted vs self-hosted models, sovereign and Indian-language options, prompt optimisation, RAG vs fine-tuning, quantisation, KV caching, batching, vLLM and inference economics.
Production service topology, durable execution, containers, AG-UI, MCP Apps, streaming progress, human approvals, exception handling and managed vs self-hosted runtimes.
Stateless services vs durable state, workers, Kubernetes pods, queues, KEDA, safe code execution, sandbox isolation, resource limits and failure recovery.
Model-aware routing, KV cache reuse and offloading, disaggregated inference, SGLang, llm-d, Dynamo, residency, availability and cost-capacity architecture.
Alongside the weekend sessions, industry leader talks and doubt-clarification sessions will be held on pre-informed weekdays.
The curriculum and schedule are subject to enhancements. A few modules, topics or their sequence may change as the programme is enriched, while retaining the core learning objectives.
At participants’ request, no sessions will be held during the Christmas and Sankranti weeks (26 December 2026 and 16 January 2027).
Free online sessions on what Forward Deployed Engineering is and how the program works.
Questions, conversations and the quiet satisfaction of making something work.

Certificate details and completion requirements will be announced with the final program release.
The final certificate title, issuer details and assessment criteria are to be confirmed.
Registration for the December program is open. Use the registration form or contact the team with questions.
For AI engineers and software engineers with roughly 2–5 years of experience, moving into customer-facing AI delivery.
Working programming skills, API familiarity and readiness to build and review software. This is a practitioner program, not a first course in coding.
Early-bird price applies to the first 20 registrations.
Total ₹1,18,000 including GST, on the standard price.
Pay by EMI on BillDeskRuns from 5 December 2026 to 14 March 2027.
Works with customers to understand a business problem and deliver a working solution, combining AI and software engineering with discovery, integration, deployment and stakeholder communication.
No prior GenAI or ML experience is needed, only solid software engineering foundations (see Eligibility). You will learn LLMs, retrieval, agents and evaluation hands-on, using a Python-first, open-source stack that includes FastAPI, Postgres, LangGraph, Qdrant, Ragas, authentik, Docker, OpenTelemetry and OpenObserve.
Yes, if your role is hands-on and you meet the prerequisites. Identity, secure integrations, data pipelines, deployment and observability are core to the curriculum.
A team capstone: an enterprise AI workflow with evidence retrieval, human approval and an audit trail, taken from discovery through deployment and handover.
No. The required labs use programme-provided scripts and free tiers. Optional experimentation outside this may cost extra.
Through working software, tests, evaluation results, architecture and security evidence, and a capstone presentation where you defend your decisions. Certificate requirements will be announced separately.
Yes, the programme is designed for working professionals. It runs for 13 weeks, including 12 weeks of lectures with 24 sessions and 72 live hours. Each session combines a 90-minute lecture with a 90-minute hands-on lab. Plan for about 9 live hours a week: 6 hours of sessions plus about 3 hours of industry lectures and doubt-clarification sessions, along with independent time for preparation and the capstone.