Systems Architect | Blockchain Infrastructure | Hardware Security | AI/ML Engineering
Building production-grade systems from embedded automotive (Bosch Engineering) to DeFi infrastructure and post-quantum cryptographic hardware. Specialized in compliant financial systems, fraud detection, multichain architecture, and secure hardware design.
Danish/Nigerian dual citizen 🇩🇰🇳🇬 | Based in Lagos | Available globally
| Repo | Stack | Status | Tested | License |
|---|---|---|---|---|
| mr-wiggum | Bash, JSON | Active | ✅ Yes | MIT |
| bio-agi | Python, LIF sim | Active | ✅ Yes | MIT |
| wallet-hardware | C++, ESP32 | Lab eval | ✅ Yes | CERN-OHL-S v2 |
| pqc-secure-element | Verilog, RISC-V | Phase 2 | CERN-OHL-S v2 | |
| auto-vision-cut | Python, MLX | Active | MIT | |
| neural-spatial-upscaler | Verilog, HLS | Design phase | ❌ No | CERN-OHL-S v2 |
| llama-server-tuning | Bash, Python | Active | ✅ Yes | GPL-3.0 |
| esp32-wireless-printserver | C, ESP-IDF | Active | MIT | |
| repo-matrix-appscript | Node.js, Apps Script | Active | MIT | |
| optimized-sssp | Rust | Active | MIT | |
| BadDocs | Python, LLM | In development | ❌ No | Private |
The fruit fly's actual connectome does the routing. Any LLM — local or remote — does the inference. No cloud required, no GPU required.
FlyWire mapped the complete wiring of an adult Drosophila brain — 138,000 neurons, 5 million synapses. bio-agi runs a real-time leaky integrate-and-fire simulation of that connectome, uses spike rates from known functional populations (mushroom body, central complex, fan-shaped body, lateral horn…) to decide which cognitive modules activate on each tick, and dispatches the activated modules to whichever LLM backend you configure.
Stack: Python, LIF simulation, SQLite
Inference backends: BitNet (CPU-native) · Ollama / any OpenAI-compat server · Anthropic SDK · dry-run mock
Cognitive modules: memory · world model · goal · reward · emotion · executive · critic gate
Built with: Mr. Wiggum autonomous loop — 12 stories, 12 commits, self-built
Use cases: on-premise email triage, personal knowledge base, interpretable security monitoring, neuropharmacology simulation
The fly brain decides what thinks. You decide with what.
A heavily extended fork of Ralph — the autonomous AI agent loop that runs AI coding tools repeatedly until all PRD items are complete.
Stack: Bash, JSON, Markdown
What this fork adds:
- Multi-tool support — Amp, Claude Code, OpenCode, Gemini CLI, or Codex; each with its own optimised prompt file
- Agent persona injection — 27 specialist personas (backend architect, security engineer, reality checker, etc.) assigned per story or whole PRD
- Aider code review gate — reviews every diff before committing; critical issues must be fixed before the loop continues
- Provider-agnostic review — review step works against local LLM, Gemini, DeepSeek, or any OpenAI-compatible API
- PATCH-not-REWRITE guards — prompt files explicitly prevent silent regressions from full file rewrites
- Live STDOUT streaming — output streamed to terminal in real time via
/dev/tty
Transforms hours of raw OBS hardware prototyping footage into clean, narrative-driven videos — entirely locally. Treats video assembly as a code execution problem driven by local multimodal intelligence.
Stack: Python, FFmpeg, MLX/mlx-vlm, MoviePy
Pipeline: FFmpeg frame extraction → local VLM scene analysis → Qwen3 script + cut-list generation → MoviePy render
Hardware: Optimised for Apple Silicon M4 Pro unified memory (64GB)
Output artifacts: event_log.json · narration_script.md · cut_list.json · output_master.mp4
Why local: hardware IP stays private; no cloud bandwidth overhead for large raw recordings
Open-source client application layer, UI state machines, and hardware abstraction layers for the TernaryCore post-quantum cryptographic reference architecture. Routes cryptographic operations through a compile-time selectable secure element HAL.
Stack: C++, Arduino/ESP32, PlatformIO
Hardware: ESP32-S3 + SSD1306 128×64 OLED + 2-button UI + ATECC608B / TernaryCore FPGA SE
SE HAL Targets: USE_SE_STUB · USE_ATECC608B · USE_TERNARYCORE_SE (FPGA UART)
Features: BIP39/32/44, PSBT signing, air-gapped operation, anti-phishing device pairing, firmware integrity attestation
⚠️ Academic/evaluation only — not certified for production key custody
Gowin GW1NR-9C FPGA implementation of the TernaryCore secure element. A PicoRV32 RISC-V soft-core runs bare-metal AT-command firmware over UART, providing ECDSA signing, secp256k1 key storage, TRNG, and SHA-256 — with a ternary polynomial multiplier for Phase 3 PQC acceleration.
Stack: Verilog, RISC-V (RV32IM), Gowin EDA, C (bare-metal)
Hardware: Sipeed Tang Nano 9K (GW1NR-LV9QN88PC6/I5)
RTL Modules: picorv32 · wb_uart · ternary_mac · ternary_poly_mul · barrett_reduce
Protocol: AT-command UART 115200 8N1 — AT+INFO · AT+RAND · AT+SIGN:ECDSA · AT+PUBKEY
Benchmarking and tuning helpers for llama-server on local inference machines. Grew out of a practical question: do alternative KV cache strategies actually improve throughput, or just reduce memory?
Stack: Bash, Python
Scripts: single/dual-endpoint benchmark · stepped one-at-a-time runner · ctx/batch/ubatch sweep · side-by-side comparison report
Published results: Apple Silicon M4 Pro — f16 vs turbo3 KV across ctx 32K–245K, with throughput and RSS data
Key finding: f16 KV remained faster at moderate context; turbo3 wins only when memory pressure becomes the bottleneck at large context windows
Firmware bridge between a modern WiFi network and a legacy wired printer, using ESP32 + W5500 Ethernet. Proxies raw TCP (port 9100), mDNS/Bonjour discovery, and IPP/AirPrint over a single low-cost microcontroller.
Stack: C, ESP-IDF, FreeRTOS, LwIP
Hardware: ESP32 + W5500 (VSPI)
Protocols: Raw TCP 9100 · mDNS · IPP/AirPrint · Captive portal config
Features: NVS config persistence, HTTP config web form, WiFi auto-reconnect with exponential backoff
Google Sheets + Apps Script dashboard that tracks repo health across GitHub (and other git hosts), with CI webhook integration and LLM-generated health summaries.
Stack: Node.js, Google Apps Script, clasp
Architecture: Node.js core engine (data fetching, analysis, multi-git-host support) + thin Apps Script display layer (Sheets UI)
Features: CI coverage webhooks · LLM health summaries · multi-provider support (GitHub, Forgejo, Gitea) · local Jest testing · Mac Mini persistent service mode
Output targets: Google Sheets · JSON · web dashboard
Research implementation of optimized Single-Source Shortest Path algorithms, progressing from classical Dijkstra toward structured paths achieving O(m log^{2/3} n) complexity.
Stack: Rust
Focus: Algorithm optimization, graph theory, benchmarking
RTL implementation of a neural spatial upscaling module targeting the Xilinx Alveo U50 HBM2 FPGA. Accelerates edge-ML inference for video/image upscaling workloads without a discrete GPU.
Stack: Verilog/SystemVerilog, HLS, Python
Hardware: Xilinx Alveo U50 (HBM2, 100G QSFP28)
Approach: Tiled systolic array with HBM2 line buffers, sub-pixel convolution, INT8 quantised weights
In development — not yet open-sourced
Intelligent documentation generation and management system. Built after experiencing too many projects with outdated or missing docs.
Stack: Python, LLM Integration, Markdown
Features: Smart content generation, quality metrics, workflow integration
Architecture and consulting work — specific implementations under NDA
- Smart Contract Architecture: Upgradeable proxy patterns (UUPS), role-based access control, regulatory compliance frameworks
- AI Fraud Detection: Predictive systems for high-transaction banking environments with real-time risk assessment
- Banking Integration: Core banking system architecture and payment processing infrastructure
- Product Stabilization: Took high-disruption products to zero critical incidents through systematic engineering
Available to discuss architecture patterns and problem-solving approaches in detail
Current Focus:
- Post-quantum cryptographic hardware (FPGA secure elements, PicoRV32 RISC-V)
- Biology-grounded AI architecture (FlyWire connectome, cognitive scaffolding)
- DeFi infrastructure and stablecoin architecture
- AI/ML fraud detection systems
- Agentic AI workflows and automation (Mr. Wiggum, Ralph pattern)
- Web3 security and smart contract auditing
Previous Work:
- Bosch Engineering: Embedded diesel powertrain systems, production automotive solutions for major OEMs
- GRIT Systems Engineering: IoT energy metering, blockchain-based P2P electricity trading research
- Digital Financial Services: Investment product platforms and payment systems architecture
Teaching & Mentorship: Alumni from teams I've built have gone on to secure pre-seed funding, publish in IEEE conferences, and receive patents for their innovations.
Blockchain & Smart Contracts:
Solidity, Hardhat, Foundry, Remix, Tendermint, Hyperledger
Web3.js, Ethers.js, OpenZeppelin
UUPS Proxies, Diamond Pattern, Access Control
AI/ML & Data:
Python, TensorFlow, PyTorch, PEFT
Fraud Detection Models, Credit Scoring
Predictive Analytics, Time Series Analysis
LIF Neural Simulation, Connectome Routing
Systems & Backend:
Go, Rust, C/C++, Python
Microservices, Event-Driven Architecture
Real-time Processing, Low-Latency Systems
Fintech & Banking:
Core Banking Systems, Banking as a Service
Payment Processing, KYC/AML Integration
NIBSS, Central Banking Aggregators
Hardware & Embedded:
ESP32, STM32, Arduino, RISC-V (PicoRV32)
Gowin / Xilinx FPGA (Verilog, Gowin EDA)
Real-time Operating Systems (RTOS)
Automotive Powertrains, IoT Devices
- Certified Blockchain Auditor — In progress
- B.Sc.Eng (Hons) — Electrical and Computer Engineering, University of Southern Denmark (2002)
- English (Fluent)
- German (Fluent)
- Danish (Fluent)
- Yoruba (Native)
- French (Conversational)
- Fractional CTO engagements for fintech and blockchain startups
- Blockchain architecture consulting for DeFi protocols and stablecoin projects
- Technical due diligence for investors evaluating blockchain projects
- Smart contract security audits
- Fraud detection system design for financial services
- Secure hardware architecture for embedded and IoT products
Not available for:
- Full-time roles requiring relocation outside Nigeria
- Projects without clear scope and timelines
- Equity-only compensation arrangements
Email: ifedayo.oladapo@gmail.com
LinkedIn: linkedin.com/in/ifedayo-oladapo
Location: Lagos, Nigeria (🇩🇰 EU work authorization available)
Response time: Usually within 24 hours for project inquiries
💡 Open to collaboration on DeFi infrastructure, fraud detection systems, blockchain education projects in Africa, open hardware security research, and biology-grounded AI.
⚡ Fun fact: I built production automotive systems before building production DeFi systems — turns out the attention to safety and reliability transfers well. Now I'm doing both simultaneously in hardware wallet silicon.