Yes, I do work full-time, but as a Generative AI consultant, an independent research blogger, and an Open Source Gen AI TDD Developer for contracts. I am in the process of upskilling and plan to start full-time work for companies by July 1st, 2026.

What training or mentoring work do I do as an Generative AI Consultant?

I train people at all skill levels to master Generative AI, whether they are students, freshers, professionals, or CXOs. Chat with me on Linkedin and book a session (once a week) at topmate.io/thomascherickal

What digital products are included in my monthly Patreon subscription?

Patreon is for publishing apps that solve pain points for this generation of job-seekers, professionals, CXOs, and even students.Join below (monthly subscription is 5 USD now, will increase as time goes by): patreon.com/thomascherickal

What digital products or playbooks do I offer on Gumroad?

I offer specialized mini-books and playbooks to master and aid in Generative AI adoption, and also add value to the buyer by leveraging their personal expertise in a novel or creative way. Many products are free, please check it out: thomascherickal.gumroad.com

What is the focus of my upcoming professional role as a Generative AI Engineer?

I plan to implement Python libraries in Rust and contribute to open-source massively with 100% code coverage and Test-Driven Development. Building on this, I pray and hope to implement research ideas like SubQ and TurboQuant and build custom SLM models that need minimal or no GPU support (only CPU inference).

Projects — Thomas Cherickal · The Digital Futurist

thomascherickal.com · The Digital Futurist

❯ ls ~/projects –lang=all –filter=genai –sort=impact

Upcoming
Open-Source
Projects

Bleeding-edge agentic systems, local LLM inference, and high-performance tooling — built for the era where AI runs on your machine, not someone else’s cloud.

12 Planned Repos

4 Languages

Gemma Primary SLM Family

100% Local-First AI

⬡ Local Inference Stack

Powered by the
Gemma Family

Every project defaults to Google’s Gemma SLM family — run via LM Studio or llama.cpp on commodity hardware (2 GB VRAM, 16 GB RAM). No API keys, no egress costs, no data leaving the machine.

Gemma 3 4B

INT4 · fits 4 GB VRAM
Primary workhorse

Gemma 4 E2B

Q4_K_M · 6 GB RAM
Reasoning tasks

Gemma 4 E4B

MoE · 2B active params
Ultra-efficient

Gemma 3n

Edge-native · 1 GB
Embedded agents

Python

Major Projects

🧠

Active

GemmaForge

A Python SDK wrapping LM Studio’s OpenAI-compatible API for the full Gemma SLM family. Supports streaming, function-calling, structured JSON output, and multi-modal prompts — a drop-in replacement for the OpenAI SDK with zero cloud dependency.

Gemma 3 / 4 LM Studio SQLite cache GitHub FastAPI Pydantic v2 asyncio

OpenAI-compatAPI surface

Local onlyInference

MITLicense

🤖

Active

AgentStack-Local

CrewAI + LangGraph multi-agent orchestration configured entirely around local Gemma inference. Agents browse the web, write and execute code, query databases, and produce verifiable outputs — zero cloud LLM calls, full DAG workflow control.

Gemma 4 E4B CrewAI LangGraph PostgreSQL GitLab CI Playwright

Multi-AgentArchitecture

DAGWorkflow

Apache 2License

🔬

WIP

QuantumCircuit-AI

Hybrid quantum-classical ML pipeline bridging Qiskit circuits with Gemma-generated circuit descriptions. SQLite caches transpiled circuits; Redis pub/sub streams partial results to a FastAPI frontend targeting IBMQ simulators and real QPU backends.

Gemma 3 4B Qiskit SQLite · Redis GitHub FastAPI PennyLane

HybridQuantum + ML

RedisStreaming

Apache 2License

Rust

Major Projects

⚡

Active

InferCore

Minimal Rust inference runtime for GGUF quantised models. Wraps llama.cpp via safe FFI bindings, exposes a streaming gRPC API, and compiles to a single static binary deployable on Raspberry Pi 5 or commodity laptops with as little as 2 GB VRAM.

GGUF Gemma llama.cpp FFI tonic gRPC SQLite session log GitLab WASM target

<5 MBBinary size

gRPCAPI

Apache 2License

🔍

Stable

GitGuard

CLI security scanner for Git repositories written in Rust. Detects secrets, leaked credentials, and policy violations across the full commit history. Plugs natively into GitHub Actions and GitLab CI; a local Gemma agent via InferCore summarises every finding in plain English.

Gemma summary GitHub Actions · GitLab CI SQLite audit log clap · git2-rs regex

Full historyScan depth

CI-nativeIntegration

GPL-3License

🌐

Research

BlockLedger-RS

Experimental Rust blockchain ledger with an embedded Gemma agent that answers natural-language queries about on-chain state. PostgreSQL stores chain data; axum serves a REST API; the WASM build target enables browser-based block verification without a server.

Gemma 4 E4B PostgreSQL axum · serde GitHub WASM sha2 · ed25519

WASMBrowser verify

NL queryChain state

Apache 2License

Minor Projects

// Golang — concise, concurrent microservices that bridge the AI stack

🔀

Stable

ProxyRouter-Go

Ultra-light Go reverse proxy that load-balances LM Studio, Ollama, and OpenRouter endpoints with per-model auth, rate limiting, and SQLite request logging — a single binary LLM gateway with zero external deps.

LLM gateway SQLite GitHub net/http · chi

Single binaryDeploy

MITLicense

📡

WIP

LogStream-AI

Structured log aggregation microservice in Go with a WebSocket endpoint where a Gemma agent delivers real-time anomaly narration for on-call engineers. Log batches persist to PostgreSQL for long-term analysis.

Gemma anomaly PostgreSQL GitLab WebSocket · zap

WebSocketLive stream

MITLicense

💬

WIP

DocuChat-Go

Zero-dependency Go CLI that ingests a directory of PDFs or Markdown files, chunks and embeds them locally, then answers natural-language questions via Gemma 4 E4B through LM Studio — private, offline RAG in a single binary.

Gemma 4 E4B LM Studio sqlite-vec GitHub RAG · embeddings

100% localInference

Single binaryDeploy

MITLicense

Mojo 🔥

Minor Projects

// Mojo — C-speed ML kernels where Python becomes the bottleneck

🔥

Research

TensorFire

Hand-written Mojo SIMD tensor kernels for GGUF dequantisation — bridges into InferCore via C ABI for 2–4× faster token generation on CPU-only systems with AVX-512 support.

GGUF Gemma kernels SIMD · AVX-512 GitHub C ABI bridge

2–4×CPU speedup

MITLicense

⚗️

Research

FlashOps

Mojo re-implementation of Flash Attention v2 for CPU targets. Reduces peak memory 40% versus naive attention, enabling longer Gemma context windows on 16 GB RAM machines without any GPU.

Gemma long-ctx Flash Attention 2 GitHub Benchmark suite

−40%Peak memory

MITLicense

📊

WIP

KernelBench-Mojo

Mojo micro-benchmark harness that stress-tests local inference kernels (matmul, softmax, RoPE) against Gemma 4 E4B layer shapes. Generates machine-readable JSON reports for easy CI regression tracking.

Gemma 4 E4B LM Studio matmul · RoPE GitHub CI JSON reports

ns-levelTiming precision

MITLicense

// Core toolchain across all projects

Git

GitHub

GitLab CI

PostgreSQL

SQLite

Redis

DuckDB

LM Studio

OpenRouter

Ollama

Docker / Podman

uv + ruff

cargo + clippy

goreleaser

llama.cpp

sqlite-vec

Thomas Cherickal Footer — Dark Mode

About the author

Thomas Cherickal

Generative AI Engineer · SLM Engineer · LLM Engineer · Generative AI Consultant · Open Source Gen AI Developer · Technical Content Writer · AI Mentor · Independent Research Blogger

thomascherickal.com · thomascherickal.github.io · Chennai, India 🇮🇳

Open for collaborations & contracts

Available for technical writing contracts, AI consulting engagements, and course collaborations. I also provide AI upskilling for individuals, AI mentoring for professionals at all levels, and special AI training for CXOs on a weekly flexible basis. Reach out via LinkedIn for a free connect, a chat, and a free consultation with a fast reply.

Find me on