AI Software Engineer at Svart App (2025-11 – Present)
Architected and developed a production-grade enterprise AI platform for digital marketing teams featuring multi-agent AI orchestration, Model Context Protocol (MCP) integrations, retrieval-augmented generation (RAG), autonomous advertising optimization, AI-powered analytics, intelligent reporting, and hyper-local keyword intelligence.
- Designed a scalable multi-agent AI platform integrating Meta Ads, Google Ads, Google Analytics 4 (GA4), Shopify, reporting, CMS, and internal business services through MCP servers, capability-based tool orchestration, semantic routing, and dynamic task planning.
- Designed and deployed a highly available AWS infrastructure using Amazon ECS, Amazon RDS PostgreSQL (pgvector), Amazon ElastiCache (Redis), Amazon S3, Amazon VPC, IAM, CloudWatch, Docker, and GPU-backed inference services, enabling horizontally scalable AI microservices, asynchronous processing, and resilient production deployments.
- Engineered a low-latency asynchronous architecture maintaining sub-100 ms API response times by keeping FastAPI request handlers lightweight (authentication, validation, request creation) while offloading retrieval, planning, MCP execution, and AI inference to Redis-backed worker services.
- Built a RAG-first execution pipeline where historical reports, campaign data, marketing documentation, and internal business knowledge are retrieved from PostgreSQL + pgvector using Qwen3-Embedding-4B (Q4_K_M) for semantic retrieval and Qwen3-Reranker-4B (Q4_K_M) for cross-encoder reranking, minimizing external LLM calls by answering previously indexed knowledge directly.
- Implemented an intelligent AI routing pipeline using Qwen3 intent classification, semantic routing, capability detection, confidence scoring, reranking, policy enforcement, and health-aware routing to determine whether requests should be rejected, answered from RAG, executed through MCP services, or escalated to DeepSeek reasoning, significantly reducing latency, token consumption, and API costs.
- Designed a hierarchical reasoning architecture where DeepSeek V4 Flash served as the primary planning and reasoning model for multi-step task decomposition, cross-platform marketing analysis, report generation, and MCP orchestration, while specialized Qwen models handled intent classification and decision verification prior to executing business-critical operations.
- Implemented durable LangGraph workflows supporting stateful agent execution, checkpointing, retries, conditional branching, parallel MCP execution, human-in-the-loop approval workflows, and fault-tolerant workflow recovery.
- Developed Python/FastAPI AI microservices integrating DeepSeek, OpenAI, Anthropic Claude, Google Gemini, LangGraph, PostgreSQL (pgvector), and RAG pipelines, providing retrieval-augmented reasoning across advertising, analytics, marketing, and internal knowledge bases.
- Fine-tuned Meta Llama and other open-source foundation models using Hugging Face with proprietary advertising datasets, improving marketing-specific reasoning, campaign recommendations, advertising analysis, and budget optimization accuracy.
- Deployed self-hosted embedding, reranking, and open-source inference services inside a private AWS VPC using containerized GPU workloads, exposing inference endpoints exclusively to backend AI services while combining proprietary cloud models for hybrid inference, improved privacy, lower latency, and reduced operational cost.
- Built an autonomous advertising optimization engine combining LLM reasoning with deterministic marketing logic (ROAS, CPA, CTR, CPC, CPM, attribution models, revenue, seasonality, campaign learning phase, budget constraints, and performance signals) to generate explainable optimization recommendations with optional human approval before execution.
- Developed asynchronous parallel execution pipelines enabling concurrent execution across advertising platforms, analytics systems, reporting services, and internal MCP servers while validating every response using Pydantic schemas before downstream reasoning, improving reliability, fault tolerance, and execution throughput.
- Implemented comprehensive observability using OpenTelemetry and Amazon CloudWatch, instrumenting distributed traces, structured logs, custom metrics, and end-to-end request correlation across FastAPI services, ECS tasks, Redis queues, LangGraph workflows, MCP servers, AI inference pipelines, and AWS infrastructure.
- Developed TypeScript/Node.js orchestration APIs that centralized prompt execution, AI workflow coordination, and backend service orchestration behind stable server-side endpoints, eliminating frontend orchestration complexity and improving maintainability.
- Built a TypeScript/Node.js headless CMS featuring dynamic schema generation, automatic CRUD endpoint generation, and Amazon S3 media storage, significantly reducing backend development effort for new content models.
- Strengthened platform security through JWT access/refresh token rotation, Redis-backed session revocation, RBAC, policy enforcement, audit logging, AI action authorization, and secure execution controls, enabling immediate server-side session invalidation and protected AI-driven business operations.
Senior Software Engineer at Instacoins (2024-01 – 2025-11)
Developed cloud-native travel booking platforms handling massive datasets and integrating multiple external APIs to facilitate secure transactions via dynamically generated cryptocurrency addresses.
- Processed 50 GB+ of payment data daily across PostgreSQL/RDS and DynamoDB paths, strategically separating transactional queries from low-latency lookups rather than forcing a single database pattern.
- Reworked crypto-to-fiat payment flows in Python/FastAPI and Golang/Gin, leveraging AWS SQS/SNS and Step Functions to enable seamless state restarts and eliminate half-updated payment entries.
- Created state-transition endpoints and worker contracts to implement a single, unified lifecycle model across all payment services, ensuring consistent status rules.
- Utilized Datadog to profile transactional paths and distributed microservice workloads, reducing latency bottlenecks and ensuring system resilience during peak data volume periods.
- Moved access-control checks directly into the payment API path using AWS Cognito MFA and Lambda@Edge request controls, positioning security checkpoints exactly where user interaction occurs.
Software Consultant / Core Team Lead at Inavitas (2023-12 – 2024-02)
Led the core engineering team to optimize overall system performance, robustness, and security, aligning technical approaches with long-term strategic product features.
- Designed high-throughput ingestion and query paths for 15M+ IoT events daily within energy-monitoring systems, optimizing paths for live telemetry, historical analytics, and operational dashboards.
- Standardized Docker/Rancher runtime configurations for over 100 microservices by automating the mapping of service URLs, exposed ports, and environment variables into repeatable config files, removing manual setup routines.
- Led a 12-person cross-functional delivery team across backend, frontend, and support, coordinating Python/Django and TypeScript/Node.js deliverables through QA handoff and release cycles.
- Established automated Jenkins pipelines and pytest gates before QA, implementing strict code reviews and pair programming to capture service configuration defects and regressions early.
Software Engineer (via Vertex Agility) at McKinsey & Company (2022-07 – 2023-04)
Backend optimization and systems modernization for McKinsey's proprietary CleanSheet product.
- Raised automated test coverage from a baseline of 4% to 75%, providing the engineering team with the safety net required to refactor code and migrate core components to AWS confidently.
- Migrated legacy backend services to AWS Lambda, Cognito, and PostgreSQL incrementally, maintaining system continuity and zero downtime by ensuring clean service boundaries.
