AI Coding Agent Development
Apna AI coding agent banao jo coding tasks aur testing mein kaam kare. Build coding agents, connect tools, automate workflows, and verify what your agents do.
Outcomes
- Build LLM applications with contracts and evidence, where the model proposes and the system enforces
- Design workflows with state, human approval, retrieval and MCP within a context budget
- Let an agent read a repository and propose reviewable patches inside a sandbox
- Evaluate agents on unseen tasks, control safety, privacy and cost, and turn the agent into a product
Projects you will build
- A tool-using LLM application
- A reviewable workflow with human approval
- RepoGuide — repository-understanding milestones
- TestFix — a controlled coding agent evaluated on twenty unseen tasks
Tools and technologies
Who this program is for
Developers and technical learners who already write code and want to engineer agent systems, not just chat with them.
Before you begin
Working JavaScript/TypeScript, asynchronous APIs, Git, terminal use and backend fundamentals. A practical readiness check precedes enrollment.
Scope
A narrow agent built on existing models. Training foundation models, unrestricted shell access and automatic production deployment are excluded, and sales are not guaranteed. Learners without JavaScript/TypeScript and backend basics should start with Full Stack Development with AI.
Laptop
Bring your own laptop (Windows, macOS or Linux, 8 GB RAM minimum; 16 GB recommended for Mobile App Development and AI Coding Agent Development).
12 modules.
300 hours.
Real projects.
Each module builds on the last and ends with practical work you do in class. Projects are reviewed by a mentor at each milestone.
Mon – Fri · 12:30 pm – 3:00 pm · 25 Nov 2026 – 11 May 2027
Module 1 · 25 hLLM Applications: Predictions, Contracts and Evidence
- Narrow task selection
- The automation spectrum
- Engineering readiness
- Model behaviour
- Token budgets
- Output variability
- Reproducibility limits
- Instruction boundaries
- Input trust
- Model-API integration
- Secret boundaries
- Latency perception
- Structured streaming
- Schema versus domain validation
- Abstention design
- Provider-failure handling
- Backoff reasoning
- Usage accounting
- Prompt versioning
- Data-conscious tracing
- Prompt experiments
- Baseline design
- Small-sample interpretation
- LLM-exercise integration
- Foundation assessment
Module 2 · 25 hTools: The Model Proposes, the System Enforces
- Tool-call lifecycle
- Tool-schema design
- Path scope
- Bounded outputs
- Result contracts
- Tool granularity
- Read-first development
- Input validation
- Tool-source trust
- Read permissions and privacy
- Step budgets
- Termination criteria
- Loop detection
- Tool concurrency
- Error feedback
- Fallback safety
- Contract testing
- Run transparency
- User-facing rationale
- Approval scope
- Approval idempotency
- Cancellation semantics
- Tool-use integration
- Capability assessment
- Module review
Module 3 · 25 hWorkflows, State and Human Approval
- Deterministic versus dynamic control
- Workflow graphs
- Structured state
- State representation
- Step granularity
- State reducers
- Persistence
- Run isolation
- Durable execution
- Replay safety
- Approval placement
- Rejection paths
- Stale approvals
- Time budgets
- Recovery classification
- Progress invariants
- Observable workflows
- Join semantics
- Fallback decisions
- Approval authorisation
- Debug artifacts
- Orchestration integration
- Recovery lab
- Assessment
- Module review
Module 4 · 25 hRetrieval, MCP and the Context Budget
- Context selection
- Retrieval choices
- Chunking trade-offs
- Evidence sufficiency
- Retrieval scores
- Source provenance
- Index freshness
- Authorised indexing
- Budget allocation
- Context compression
- Protocol versus trust
- MCP concepts
- Capability enablement
- Connection boundaries
- Audience-bound credentials
- Capability-change review
- Indirect prompt injection
- Layered defence
- Missing-evidence behaviour
- Evidence precedence
- Retrieval evaluation
- Integration exercise
- Optimisation lab
- Assessment
- Module review
Module 5 · 25 hUnderstand the Repository: From File Tree to Behaviour
- Repository orientation
- Untrusted project scripts
- Project detection
- Dependency reproducibility
- Execution entrypoints
- Call-site analysis
- Type dependencies
- Failure attribution
- Clarification
- Specification sources
- Search strategy
- Scope exclusions
- Sensitive-context handling
- Snapshot consistency
- Parsing trade-offs
- Scope discipline
- Test-driven investigation
- Environment dependencies
- Stack interpretation
- Task-specific summaries
- Evidence packaging
- Repository assistants
- Generalisation checks
- Investigation assessment
- Module review
Module 6 · 25 hPropose a Patch Without Taking Over
- Bounded action plans
- Minimal patches
- Patch provenance
- Mutation scope
- Patch validation
- Protected evaluation assets
- Dependency changes
- Behaviour-preserving constraints
- Untrusted directives
- Change classification
- Reviewable artifacts
- Content-bound approvals
- Atomic patching strategies
- Protecting user changes
- Scoped recovery
- Evidence-based summaries
- Alternative evaluation
- Search budgets
- Feedback loops
- Replanning boundaries
- Mini-project extension
- Patch critique
- Concurrency lab
- Assessment
- Module review
Module 7 · 25 hSandboxing: Don't Give the Model the Keys to the Machine
- Isolation threat models
- Execution capabilities
- Structured execution
- Filesystem scope
- Egress policy
- Controlled provisioning
- Credential minimisation
- Resource limits
- Memory quotas
- Storage quotas
- Output limits
- Process lifecycle
- Least-privilege execution
- Execution identity
- Immutable inputs
- Tenant isolation
- Artifact-export validation
- Execution-outcome semantics
- Trusted test harnesses
- Safe operational logs
- Execution APIs
- Execution integration
- Concurrency lab
- Assessment
- Safety review
Module 8 · 25 hThe Test-Fix Loop, Jobs and Review Dashboard
- Failure-feedback packaging
- Bounded repair loops
- Attempt isolation
- Independent tests
- Regression comparison
- Metric integrity
- Nondeterminism analysis
- Job architecture
- Job identity
- Job-state transitions
- Failure recovery
- Race handling
- Status delivery
- Run observability
- Outcome reporting
- Review experience
- Reviewer permissions
- Action boundaries
- Artifact authorisation
- Retention choices
- Product integration
- Fault isolation
- Small-load validation
- Assessment
- Integration review
Module 9 · 25 hEvaluation: Turn the Demo into Evidence
- Success criteria
- Multidimensional scoring
- Fair baselines
- Dataset separation
- Evaluation composition
- Data leakage
- Metric arithmetic
- Attempt accounting
- Evaluation independence
- Specification alignment
- Rubric calibration
- Failure taxonomy
- Coverage versus precision
- Economic evaluation
- Distribution awareness
- Uncertainty fundamentals
- Experiment control
- Controlled experiments
- Component-value tests
- External validity
- Learning from failures
- Reproducible reporting
- Metric critique
- Assessment
- Module review
Module 10 · 25 hSafety, Privacy, Quotas and Unit Economics
- Threat modelling
- Layered mitigation
- Safe adversarial fixtures
- Capability-escalation checks
- Memory isolation
- Retention policy
- Deletion propagation
- Policy composition
- Atomic quota reservation
- Budget uncertainty
- Usage reconciliation
- Cost-quality trade-offs
- Total operating cost
- Usage-based product design
- Trial controls
- Idempotent billing demonstrations
- Eventual-consistency experience
- Operational judgement
- Emergency controls
- Safe incident response
- Continuous review
- Failure-cost lab
- Isolation lab
- Assessment
- Module review
Module 11 · 25 hProductisation: From Agent to Usable Product
- Customer selection
- Customer discovery
- Demand evidence
- Trust and onboarding
- First-use experience
- Onboarding diagnostics
- Compatibility promises
- Interaction design
- Actionable status experience
- Decision support
- Designing for failure
- Accurate product messaging
- Pilot planning
- Consent and authority
- Confidentiality
- Commercial scoping
- Product positioning
- Support operations
- Product deployment
- Operational configuration
- Learning metrics
- Product storytelling
- Integration lab
- Assessment
- Release review
Module 12 · 25 hCapstone: Unseen Tasks, Honest Results and Handover
- Evaluation freeze
- Evaluation governance
- Fair evaluation
- Safety readiness
- Reproducibility
- Held-out execution 1
- Evaluation observation 1
- Held-out execution 2
- Evaluation observation 2
- Held-out execution 3
- Evaluation observation 3
- Held-out execution 4
- Evaluation observation 4
- Held-out execution 5
- Evaluation finalisation
- Final metric reporting
- Final control assessment
- Final approval assessment
- Final reliability assessment
- Post-evaluation transfer task
- Delivery artifacts
- Pilot-proposal defence
- Capstone defence A
- Capstone defence B
- Completion and next steps
What can you earn after this course?
₹5–8 LPA is a common fresher range for AI engineers; GenAI and agent skills are reported to earn a 25–45% premium.
Entry-level AI / LLM engineer job
₹5–8 LPA common · up to ₹10–18 LPA reported for 0–2 years at AI-first companiesGlobal capability centres and AI-native startups are reported to pay 40–70% more than IT services.
- AI / LLM application engineer
- AI automation developer
- Agent and evaluation engineer
- Freelance AI chatbot builder
Basic FAQ / support chatbot
from about ₹35,000 per buildVendor-published low end for small businesses.
GPT-style assistant on a business's documents
₹3–8 lakh per build (agency pricing)Freelancers usually quote less; start with smaller, scoped builds.
Managed chatbot for a small business
₹5,000–₹25,000 a monthRecurring income for support and updates.
Example: Two small businesses paying ₹10,000 a month each for a maintained support assistant is ₹20,000 a month in recurring income, on top of a job.
These are published 2026 market ranges for India, not a promise. What you earn depends on your skill, portfolio, practice, city, employer and the effort you put into finding work. No job, client or income is guaranteed.
How to start earning — full guideFounding batch
- First class
- 25 Nov 2026
- Ends
- 11 May 2027 (End dates assume classes every weekday; public holidays may move the last class later.)
- Days
- Mon – Fri
- Time
- 12:30 pm – 3:00 pm
- Teaching & lab hours
- 12.5 h/week · 300 h total · 120 classes
- Home practice
- 8-12 hours/week
- Format
- Offline classroom
- Venue
- Udaipur, Rajasthan · offline classroom
- Language
- Hinglish (Hindi + English)
- Batch size
- Up to 15 learners
- Regular fee
- ₹80,000
- Offer fee
- ₹58,000 (Navratri & Diwali offer · valid till 8 Nov 2026)
- Pre-booking
- ₹2,000 token, adjusted in instalment 1
₹58,000 ₹80,000Save ₹22,000
Navratri & Diwali offer · valid till 8 Nov 2026. Pay in full or in instalments; the total is the same either way. Pre-book your seat with ₹2,000, adjusted in instalment 1.
- Instalment 1 · At admission
- ₹13,000
- Instalment 2 · 25 Dec 2026
- ₹9,000
- Instalment 3 · 25 Jan 2027
- ₹9,000
- Instalment 4 · 25 Feb 2027
- ₹9,000
- Instalment 5 · 25 Mar 2027
- ₹9,000
- Instalment 6 · 25 Apr 2027
- ₹9,000
How the token works: total fee ₹58,000. Pay ₹2,000 now to reserve a seat; at admission pay ₹11,000 more to complete instalment 1 (₹13,000). The total stays ₹58,000.
The final payable amount and invoice details are confirmed in writing before you pay. No registration, assessment or certificate charges are added later.
Pre-book seat · ₹2,000Included in the fee
- Classroom instruction and scheduled guided labs
- Course material and project starter resources
- Defined project reviews and doubt support
- Assignments, assessment and a completion certificate
- A controlled allowance for required classroom AI/API exercises
- Portfolio and demo-day preparation
Support
Scheduled doubt support during the course, milestone-based project reviews, and 30 days of post-course project clarification.
Not included
- Personal premium software subscriptions
- A custom domain you choose to own
- Hosting after the course ends
- Personal app-store developer accounts
- Personal API usage beyond the included allowance
- Your own advertising spend outside class exercises
Free demo class
Inspect and validate an agent-generated code change in one session.
Book a free demo