# Daniel Plas Rivera — morphus56k.com

> Lead Software Architect at McKinsey & Company (New York). 15+ years across
> GenAI/LLM platform engineering, data engineering at billions-of-records scale,
> cloud infrastructure, and low-level systems reverse engineering. This site is
> his technical portfolio and research notebook.

Key facts:
- AI/LLM platforms: built a Model Context Protocol (MCP) server exposing 75+ agent
  tools behind a tool-RAG discovery layer; a multi-provider LLM gateway (Anthropic,
  OpenAI/Azure, Google/Vertex, AWS Bedrock) with SSE streaming and citation
  grounding; and a fine-tuned hierarchical DeBERTa classifier (4-level, 137-path
  taxonomy, deepest-level accuracy 4%->77%+) with SageMaker MLOps — serving 25+
  McKinsey teams in production.
- Data engineering: AWS Glue/Spark/Snowflake pipelines; re-architected 16.5M daily
  API calls to sub-million on Kinesis; 99.6% infrastructure cost reduction.
- Systems research: CPS-2 arcade reverse engineering (68000 assembly, custom
  FBNeo emulator cores, 64MB graphics cap-break) and a memorial fighting-game AI
  (VQ-BeT behavioral cloning + style-constrained PPO, trained on Apple MLX).
- Previously: Capital One, PwC.

## Pages

- [Live demo](https://morphus56k.com/demo): "Spar with the Ghost" — the 2.2M-parameter memorial
  policy (trained on 7,554 recorded matches) running live in the browser via a
  hand-written forward pass; original abstract arena, no game assets
- [Resume](https://morphus56k.com/resume): canonical resume, senior/staff register
- [About](https://morphus56k.com/about): full service record, skills, awards
- [Post index (JSON)](https://morphus56k.com/api/posts.json): typed index of all posts
- [Profile (JSON)](https://morphus56k.com/api/profile.json): machine-readable profile

## Research series

- Production GenAI
- CPS-2 Reverse Engineering

## Posts (markdown mirrors)

- [What Does a Trustworthy AI Analyst UI Actually Look Like?](https://morphus56k.com/md/trustworthy-ai-frontend.md): Lessons from shipping a client-ready React 19/TypeScript front end for an AI analyst: SSE streaming with click-to-source citations, constrai
- [How Do You Lift a Classifier's Hardest Level From 4% to 77%?](https://morphus56k.com/md/hierarchical-classifier-imbalance.md): Fine-tuning a hierarchy of DeBERTa/RoBERTa-family transformers to route records through a 4-level, 137-path taxonomy — and how class-weighte
- [How Do You Pick Which Model to Ship?](https://morphus56k.com/md/llm-eval-harness.md): Building an offline eval harness that compares frontier and baseline models on latency, token cost, and output quality — with automated hall
- [How Do You Make an LLM Cite Its Sources — and Prove It?](https://morphus56k.com/md/grounded-llm-citations.md): Provider-native citation grounding as an explainable-AI trust layer: page-anchored source quotes rendered as click-to-source footnotes, clai
- [How Do You Give an AI Agent 75 Tools Without Wrecking Its Judgment?](https://morphus56k.com/md/agent-tool-rag-discovery.md): A tool-RAG discovery surface — five verbs instead of seventy-five schemas — that lets an agent find, understand, and safely compose a large 
- [Production GenAI Series](https://morphus56k.com/md/production-genai-series-index.md): Complete series index: engineering LLM platforms that survive production — agent tool discovery at scale, grounded citations, eval-driven mo
- [Can You Bring a Fightcade Player Back? The Complete Memorial Bot Story](https://morphus56k.com/md/memorial-bot-complete-arc.md): The definitive writeup of the alanmargolies88 memorial AI: mining 7,554 recorded matches, reverse-engineering CPS-2 RAM until the bot could 
- [Cabinet-Faithful Spectating: Why You Can't Just memcpy a Game State Across Architectures](https://morphus56k.com/md/cabinet-parity-spectate.md): The sequel to the Arcade1Up protocol RE: disassembling libarcade.so's spectate consumer, discovering the cabinet's blind-apply-and-run model
- [Reverse Engineering Arcade1Up's Online Fighting Protocol](https://morphus56k.com/md/arcade1up-online-reverse-engineering.md): From TCP packet captures to P2P input exchange at 60fps — reverse engineering Arcade1Up's proprietary online multiplayer system for Marvel v
- [How Do You Add a New Secret Character to a 1998 Fighting Game?](https://morphus56k.com/md/armored-spiderman-hidden-box.md): Reverse engineering MvC1's hidden character box system - discovering three table formats, cursor navigation mechanics, and the path to makin
- [Binary Forensics: Recovering an Unreleased Character from a 1998 Fighting Game](https://morphus56k.com/md/binary-forensics-unreleased-character.md): Systematic binary analysis discovers 85 code references, 4 disabled load points, and 3 distinct table formats for character ID 0x2E -- evide
- [What Are the Absolute Hardware Limits of CPS-2 Arcade Boards?](https://morphus56k.com/md/cps2-maximum-rom-expansion.md): Pushing every ROM type to the theoretical maximum of Capcom's CPS-2 hardware: 128MB graphics, 16MB audio, 4MB program - documented with in-g
- [CPS-2 Reverse Engineering Series](https://morphus56k.com/md/cps2-series-index.md): Complete series index: reverse engineering Capcom's CPS-2 arcade hardware from first principles -- binary forensics, ROM expansion, audio in
- [How Do You Add a Custom Diagnostic Screen to Arcade Hardware?](https://morphus56k.com/md/cps2-test-menu-engineering.md): Engineering a new menu item and 24-line verification screen into CPS-2 arcade firmware using 68000 assembly, jump table extension, and sprit
- [Can You Port a Dreamcast Soundtrack to a 1998 Arcade Board?](https://morphus56k.com/md/mvsc2-soundtrack-port.md): Porting MvC2's jazz-fusion soundtrack to MvC1's CPS-2 hardware - from 8-bit PCM injection to multi-bank streaming playback via Z80 ISR-drive
- [Selected Work](https://morphus56k.com/md/selected-work.md): Curated engineering case studies: production GenAI/LLM platform engineering, a memorial fighting-game AI, and binary-level systems engineeri
- [What Would a Next-Gen Emulator Look Like?](https://morphus56k.com/md/ai-augmented-developer-tools.md): Designing an emulator platform with Metal GPU rendering, AI opponents via CoreML/PyTorch, massively parallel training, and full developer to
- [Binary Forensics: What Hidden Features Exist in Compiled Binaries?](https://morphus56k.com/md/cryptographic-pattern-analysis.md): Discovering undocumented characters, secret input codes, and unused entity IDs through systematic binary forensics on MvC1's CPS-2 ROM
- [Can You Max Out a 1998 Arcade Board's Storage?](https://morphus56k.com/md/memory-management-binary-expansion.md): Expanding every ROM type on CPS-2 hardware to its theoretical maximum - 128MB graphics, 16MB audio, 4MB program - using Python automation an
- [How Do You Make Reverse Engineering Work Reproducible?](https://morphus56k.com/md/reproducible-analysis-workflows.md): Structuring technical analysis so both humans and AI can execute it - from scattered ROM notes to automated validation pipelines with in-gam
- [What's Really Happening Inside a Binary's UI State Machine?](https://morphus56k.com/md/reverse-engineering-entity-systems.md): Reverse engineering MvC1's character select screen - discovering cursor navigation tables, decoding entity ID mappings, and mapping three di
- [Can You Inject New Behaviors into a Binary Without Source Code?](https://morphus56k.com/md/state-machine-injection.md): Adding a new secret character to a 1998 arcade game through ROM patching, 68000 assembly injection, palette engineering, and input sequence