Biresh Biswas

Lightning Network·Rust·LDK·Bitcoin·gRPC
•
Selected to the highly competitive open-source program; contributing to Stable Channels – a
Bitcoin-collateralized stability protocol – and upstream LDK Server, the official Rust Lightning daemon
maintained by lightningdevkit.
•
Working across the full Lightning stack – protocol logic, server architecture, gRPC API surface, and Lightning
Network primitives such as channels, HTLCs, and custom TLV records – to harden and extend production-grade
Bitcoin infrastructure.
•
Driving upstream open-source contributions into the canonical LDK Server repository – closing API gaps and
bringing new capabilities into the mainline daemon used across the Lightning ecosystem.
•
Working toward broader operator decentralization on Lightning, making it easier for individuals and small
businesses to self-host the financial infrastructure that powers real-world Bitcoin payments.

DeFi·Multi-Chain·Agentic Systems·Python
•
Built and open-sourced defi-skills – a deterministic engine that converts natural-language intents into unsigned,
auditable on-chain transactions across 13 DeFi protocols and 6 chains.
•
Architected a JSON-playbook system where adding a new protocol requires zero engine code– resolvers, ABI
encoding, approvals, and multi-chain addresses are fully data-driven, lowering contributor friction to a single PR.
•
Shipped multi-platform distribution (Claude Code plugin, OpenClaw skill, hosted REST API, pip package)
with security hardening, structured error contracts, and reproducible benchmarks.
•
Drove early adoption – 7 contributors merging PRs, 200+ monthly PyPI downloads, and 20K+ launch-post
reach within weeks of release.

Distributed Training·PyTorch·DeepSpeed·Hugging Face PEFT
•
Designed SplitLoRA– a distributed fine-tuning framework for edge device networks combining layer-wise
model splitting with LoRA adapter allocation to coordinate training across heterogeneous
CPU/GPU/multi-device topologies.
•
Built the framework on DeepSpeed and FSDP to shard model weights and optimizer states across constrained
devices; benchmarked scalability, throughput, and downstream accuracy trade-offs across device configurations.

Docker·Node.js·GCP·CLI Workflows
•
Reduced production Docker image size from 10 GB to 1 GB (∼90% reduction) via multi-stage builds, engineering
a fully headless PrusaSlicer CLI workflow deployable on GCP VMs.
•
Integrated the slicing engine with the backend for automated dynamic pricing and metadata extraction from
G-code, taking the pipeline from prototype to production.

ASP.NET Core·SQL Server·QuestPDF·Azure AI
•
Built FlowDocs– an enterprise documentation platform with Azure AI-powered draft generation, cutting
manual authoring effort by 60% and review cycles by 45%.
•
Implemented a role-based approval chain (contributor→team lead→department head) with semantic
versioning and audit trails for compliance traceability, on a normalized SQL Server schema with EF Core and
QuestPDF.