“Trust Whole-Heartedly In the Lord, Put No Faith In Your Own Understanding, Acknowledge Him In Every Step You take, and He Will See That Your Paths Are Smooth.”

Proverbs 3:5-7

Coding Profile & Engineering Methodology

An architectural breakdown of my coding habits, technology stack, and daily workflows for building SOTA (State of the Art) systems and custom projects.

---

🏗️ Core Engineering Methodologies

When architecting new solutions, my coding habits follow strict, modern paradigms. Modern software engineering requires a departure from traditional, manual keystrokes, moving instead toward a synthesis of high-level architectural planning and intelligent orchestration.

AI-First Orchestration:

• I do not code in isolation, nor do I rely solely on traditional IDEs.
• I use Google Antigravity as my central AI orchestration tool for absolutely every task.
• This encompasses the entire software development lifecycle.
• With Antigravity, I directly scaffold thousands of lines of code, construct boilerplate, and wire complex API integrations in seconds.
• But its utility extends far beyond programming; I use it to automate tedious non-coding administrative work, streamline daily project management sprints, and orchestrate automated Git versioning flows.
• This complete orchestration allows me to function not just as an engineer, but as an entire development studio—focused on high-level systems design rather than syntax.

Test-Driven & Verifiable:

• In an era of AI-generated code, verification is the most critical step of development.
• Every line of code I generate or write is meticulously tested.
• I natively test all functional code blocks and algorithms directly within Google Antigravity’s execution environment before deployment, ensuring logical soundness and runtime stability.
• However, code execution is only half the battle.
• For architectural fact-checking, mathematical proofs, and validating edge-case logic, I cross-reference my systems heavily.
• I utilize Perplexity.ai for real-time technical research and documentation deep-dives, while employing Claude Sonnet for rapid, iterative reasoning, and Claude Opus for solving extremely complex, multi-layered architectural challenges that require deep contextual awareness.

Advanced Prompt-Driven Development:

• Much of my engineering stems directly from advanced prompt engineering rather than manual typing.
• I recognize that natural language is the new assembly language.
• I utilize explicit Chain of Thought prompting to force models to break down complex state machines into logical, step-by-step reasoning paths.
• I use Role Strategizing to constrain the AI’s perspective—forcing it to review my code strictly as a “Senior Security Auditor” or a “Staff Systems Architect.”
• Furthermore, I employ Multiple Prompt Refinement, systematically feeding outputs back into the context window to iteratively optimize performance and eliminate hallucinations.

---

🚀 The Technology Stack

🥇 Primary Tier: Ease of Use, Performance, AI, Blockchain, and Quantum combined!

• Python: The undisputed, foundational backbone of my machine learning and LLM engineering. I heavily utilize Python for exploratory data science, rapid prototyping, neural network implementations, and building complex AI agent frameworks. Its massive ecosystem of libraries (like PyTorch and TensorFlow) and the interactive nature of Jupyter environments make it essential for testing algorithms, plotting data, and visualizing tensor operations before pushing models to production.

• Golang: My deployment and orchestration layer for shipping AI services that actually survive contact with production. I write Go for containerized microservices, REST and gRPC API gateways that front my model endpoints, and the CLI tooling that glues my pipelines together. Goroutines and channels give me cheap, readable concurrency for handling thousands of simultaneous inference requests, while the standard library and fast compile times keep iteration speed close to Python’s without giving up runtime performance.

• Rust: My go-to language for building high-concurrency, memory-safe AI infrastructure and data processing pipelines. While Python excels at high-level logic, I leverage Rust to develop the critical “plumbing”—such as high-performance vector database connectors and custom tokenizers—where memory safety and thread-level parallelism are non-negotiable. Its zero-cost abstractions and “fearless concurrency” ensure that my underlying systems remain stable and lightning-fast under heavy production loads.

🥈 Secondary Tier: Web Architecture, Version Control & Systems Glue

Version Control & CI/CD: Relentless, meticulous use of Git and GitHub — and not only for code. I version my system prompts, datasets, evaluation harnesses, and architectural markdown, because a system prompt is a production artifact with its own regressions. Atomic semantic commits, version-controlled pre-commit hooks, GitHub Actions for linting, tests, builds and deploys, protected branches, tagged releases.

Web Fundamentals: HTML, CSS, and JavaScript for responsive interfaces and interactive dashboards. Semantic markup, modern CSS (Grid, Flexbox, custom properties), React with Tailwind only when the interface warrants it. Static sites on GitHub Pages: free hosting, Git-native deploys, CDN delivery, no server to patch. I use AI most heavily here, and find it reliable — my weapon of choice is Google Antigravity.

Linux & CLI: Zorin OS daily driver. Shell scripting, systemd, cron, SSH, and the grep/sed/awk/jq chain for data that has not yet earned a Python script.

Containers: Docker and multi-stage builds for small, reproducible images; Kubernetes when a workload needs real scheduling. Go’s static binaries make distroless images measured in megabytes rather than gigabytes.

Data & Docs: PostgreSQL and SQLite; pgvector and Qdrant for retrieval; DuckDB and Parquet for local analytics. Markdown for humans, JSON-LD and Schema.org for machines, Mermaid diagrams versioned alongside the code.

---

🤖 AI Tools and Environments

I have moved away from fragmented coding assistants and paid cloud platforms toward unified, autonomous intelligence. Autocomplete was never the product; autonomy is.

The core triad. Antigravity is the orchestration cockpit — project scaffolding, multi-file refactors, API wiring, Git automation, and the admin work that eats an engineer’s week. Claude Code is the terminal-native agent: repository-aware, suited to surgical work in existing codebases and to maintaining versioned skills that encode my conventions. Google AI Studio is the model laboratory for prompt iteration and behaviour comparison before anything reaches production.

Verification. Generation is easy; verification is the job. Perplexity for source-linked research, Claude Sonnet for rapid iteration, Claude Opus for complex architectural problems, Gemini for long-document analysis. Cross-model agreement raises confidence; disagreement locates the real problem.

Local & open-weight. Ollama and LM Studio for management, llama.cpp with GGUF quantization for modest hardware, vLLM for throughput, Hugging Face for weights. My local model changes monthly — that is the state of the field, not indecision. Privacy, cost, and offline capability are why it matters.

Agent plumbing. MCP servers for governed tool and filesystem access, A2A for handoff, custom Python where a framework would hide the control flow.

---

🎨 Visual Architecture & Web Design

Backend systems are my focus, but I have a good eye for design — and a system nobody can navigate is a system nobody adopts.

Principles. Hierarchy before decoration. Typography carries most of the design: two typefaces, a real scale, capped line length. A dark register with accent colour used as punctuation, not wallpaper. Whitespace is a feature. Mobile-first, always.

Platforms. WordPress for scalable content management, plugin depth, and structured data — my best experience overall. Framer for pixel-perfect interactive design. Wix for fast landing pages with built-in form handling. SITE123 when speed and round-the-clock support outrank features.

Hand-rolled when it matters. My GitHub Pages site is hand-written HTML and CSS, versioned in Git and CDN-served for free. Platform when velocity wins; hand-rolled when control wins.

AI-assisted loop. Generate several genuinely different directions, discard anything templated, refine the survivor across multiple prompt rounds, test in a real browser. The model produces the pixels; the judgement about which pixels deserve to exist remains mine.

Non-negotiables. Lighthouse and Core Web Vitals as pass/fail gates. Semantic landmarks, real alt text, keyboard navigation, WCAG AA contrast. JSON-LD, Open Graph, canonical URLs, accurate sitemap. Both halves of the practice are live at thomascherickal.com and thomascherickal.github.io.

However, my best experience has been with WordPress; a little learning goes a long way!

I use AI heavily, but I use open-source models and/or Google tools to ensure I do not have to pay for them!

My local LLM of choice changes every month!