Berk Delibalta · Software & AI Engineer
Pioneering at the frontier of engineering.
I earned my M.S. in Computer Science from Boston University and bring more than five years of experience across enterprise software, cloud infrastructure, and applied AI. My work spans deep reinforcement learning and simulation, production backend systems, and AI-enabled products for complex operational environments.
- AI & research
- Deep reinforcement learning, intelligent agents, continuous control, simulation
- Engineering
- C++, Java, Python, backend systems, cloud infrastructure, production AI
- Applications
- Financial, industrial, and other complex enterprise systems
01
Research & open source
An open-source C++ implementation of Twin Delayed Deep Deterministic Policy Gradient for continuous-control experiments, with MuJoCo simulation and Eigen and LibTorch execution paths.
STATEst
→
ACTORπ(s)
⇢
Q1(s,a)Q2(s,a)
→
UPDATEmin Q′
targety = r + γ min(Q′₁, Q′₂)Twin critics reduce overestimation; actor updates are delayed.
- Implemented TD3 in modern C++ with replay buffers, normalization, training utilities, and CMake builds.
- Connected the learning loop to MuJoCo continuous-control environments.
- Built separate Eigen and LibTorch paths to support systems-level experimentation.
- Published a stable working release with automated build and test infrastructure.
Stack: C++17, CMake, MuJoCo, Eigen, LibTorch
Boston University · Research engineering
High-performance RL experimentation
Contributed to a research effort exploring reinforcement-learning execution in C++ and Python. My work focused on the difficult integration layer: adapting MuJoCo for the C++ framework, enabling algorithms to run against multiple simulated agents, and working directly with TD3 and PPO experiments.
This was research engineering work under university supervision; no claim of publication authorship is made here.
02
Applied systems
A bid-operations system for industrial teams managing RFQs, changing drawings, amendments, requirements, owners, clarifications, and submission deadlines.
01IngestDocuments and revisions
02StructureRequirements and ownership
03ControlIssues, audit, readiness
I translated the operational workflow into a multi-tenant product with structured ingestion, permissions, auditable AI actions, issue tracking, and submission-readiness controls. The first release is approaching completion.
Stack: Python, FastAPI, TypeScript, PostgreSQL, RBAC, LLM workflows
Deloitte · Enterprise integration
Boomi API request pacing
Designed batching and paced dispatch for a Salesforce-facing workflow that needed to move large order sets through a constrained Boomi API without overwhelming the upstream integration.
Stack: Node.js, Express, Boomi, Salesforce, REST APIs
03
Experience
Fitzroy / Maria AnnFounder & hands-on engineer
Product discovery, architecture, backend systems, cloud infrastructure, and applied AI.
DeloitteCloud & software engineering
Enterprise integrations, serverless systems, API reliability, delivery, and production troubleshooting.
IBMCloud platform engineering
Backend functionality, runtime images, and Kubernetes deployment for IBM Cloud Box work.
Zurich · ThalesSoftware engineering
Java enterprise applications, production support, and engineering work in regulated environments.
04
Education & credentials
- M.S. in Computer ScienceBoston University
- Computer EngineeringPolitecnico di Torino
- AWS Certified Solutions ArchitectCloud architecture credential
Turin→Sophia Antipolis→Boston