{ Filip Novakov }

Key Projects {

IBM (

[Net Zero Cloud Implementation]

// 6 monts duration

As the only team member allocated 100% to the project on my team I was responsible for most of the configuration and development of the Salesforce platform and Net Zero Cloud (NZC) including:

• Activating the functionalities required for proper functioning of the Salesforce platform and NZC
• Designing the client's company hierarchy in Salesforce
• Designing user profiles, roles and levels of access as per client requirements
• Customizing the standard objects of NZC and Salesforce to support client's data model
• Adapting the page layouts of the NZC app for client's use case
• Creating templates for manual upload of data to NZC
• Mapping the integrated systems data model to NZC data model

[Agentforce Pilot Project]

// 6 weeks duration

Agile project with daily stand-ups focused on rapid prototyping and fast iteration to deliver a working POC. The challenge was streamlining and automating much of the cumbersome process of marketing campaign creations. A custom agent was built which interacted with Watsonx platform through Apex classes and used flows to generate the campaign copy personalized to a given customer segment for 2 different channels.

)

Dynamous Kiro Hackathon (

[Iubar: AI-Enhanced Knowledge Management Platform]

// January 2026, 68 hours

Built a hybrid-RAG personal knowledge management web app (FastAPI + React/TypeScript) combining ChromaDB vector search, SQLite structured memory, Voyage AI embeddings, and DeepSeek V3.2-Exp to turn uploaded documents and GitHub repos into a chat-based, source-attributed AI research assistant.

AI application engineering:

• Designed the hybrid-RAG ingestion pipeline: Docling/gitingest document conversion, token-aware chunking (512-1024 tokens, 15% overlap), and Voyage AI embeddings feeding a ChromaDB vector store
• Integrated DeepSeek V3.2-Exp for streaming, source-attributed chat responses, with context caching (~90% cost reduction on repeated queries), LRU response caching, and per-session rate/spend limits
• Added production-hardening measures: prompt-injection defenses, a circuit-breaker pattern for API resilience, and structured logging, validated by a 116+ test suite across backend, frontend, and E2E coverage

Agentic coding process:

• Authored 13 custom prompt commands (context priming, complexity-scored planning, checkpointed execution, code review) driving a spec-first workflow from requirement to tested code
• Configured 4 scoped sub-agents (backend, frontend, review, UX) with role-specific tool permissions and documentation resources for isolated, parallel agent work
• Built 8 automation hooks (auto-formatting, Playwright UX validation, Postman API sync) and tiered steering rules (always-on safety rules vs. manually-loaded docs) to keep agents in-scope and control context/token usage
• Designed a checkpointed state-tracking system so agents could resume interrupted multi-step tasks without losing progress or re-deriving context

)

GreenFox (

[Foxbuy: E-bay clone (teamwork)]

// 6 weeks duration

Simulated work environment (incl. daily standups, JIRA agile development, application design and technical diagramming)

Main challenges: Researching and implementing the project in real time using unfamiliar technologies

Technologies used: Spring Boot, Spring Security, GitHub, Miro board, Jira Agile board, Junit, MockMVC, Swagger, Jackson, JWT, JaCoCo, Flyway, MySQL, Mailtrap

)

}