Skip to content

AI Tools & LLM

Focused AI tools — the thinking done where it counts.

Meng Tech builds focused AI features — retrieval and embeddings, chat, agents, and AI coaches — and wires LLMs into real products with human-in-the-loop workflows. AI where it earns its keep, not AI for its own sake.

What this covers

Model-shaped features, wired into real products

The useful, unglamorous AI work: grounded answers, agents that do something, and features that hold up in production.

  • Embeddings & RAG

    Retrieval over your own documents and data, so answers are grounded in what's true for your business.

  • Chat & agents

    Conversational interfaces and tool-using agents that actually do work, not just talk about it.

  • LLM-shaped features

    Summarize, classify, extract, draft — the model-shaped parts of a product, built to be reliable.

  • AI coaches

    Guided, domain-aware assistants that walk a user through something, with a point of view.

  • Model integration

    Wiring LLMs into an existing product cleanly — behind a sensible API, with cost and latency in mind.

  • Human-in-the-loop workflows

    Review, approval, and correction designed in — so people stay in control where it matters.

Who it's for

Built for teams with a real place for AI

The teams we do our best AI work with usually arrive in one of these situations — with a genuine problem for a model to solve, not a mandate to 'add AI'.

  • Product teams adding AI features

    You've got a product and a place where AI would genuinely help. We build the model-shaped feature and wire it in cleanly, without derailing your roadmap.

  • Teams sitting on a pile of documents

    Contracts, tickets, knowledge bases, records. We build retrieval over your own data so people can ask it questions and get answers grounded in what's actually true for you.

  • Businesses figuring out where AI fits

    You know AI matters but not where it earns its place. We'll be honest about where it helps and where it doesn't — and build the part that does.

Tools we've built

AI we've shipped, end to end

A few of the AI products we've built ourselves — the same engineering we'd bring to your model-shaped problem.

Decision Engine

TFT HexSight

A solver for Teamfight Tactics — the way a poker solver or chess engine works, but for TFT. It evaluates a game's state and computes the higher-EV line: when to roll or level, how to build and position. Post-game analysis today, fully within Riot's Terms.

Explore
Structured Extraction

Ironform

Point it at a messy invoice, email, or contract and get back clean, schema-perfect data — every field typed and validated, never a malformed answer to babysit.

Explore
Audio → Notes

Scribewave

Drop in a recording and get a tidy, searchable transcript and summary — running on your own machine, so nothing sensitive ever leaves the room.

Explore

Pengu and Teamfight Tactics are trademarks or registered trademarks of Riot Games, Inc. TFT HexSight isn't endorsed by Riot Games and doesn't reflect the views or opinions of Riot Games or anyone officially involved in producing or managing Riot Games properties.

From our lineup

AI products of our own

Beyond client work, we build AI-driven products under our own name — the ones from our lineup where the model does the heavy lifting.

  • On-Device AI

    Cutaway

    An AI image studio where nothing gets uploaded — background removal runs entirely in the browser, with live edge refinement and instant scene swaps.

  • AI Imaging

    Restyle

    A photo restyler that turns a snapshot into a hand-painted film still — pick from a set of painterly presets, slide between before and after, and download the finished frame.

  • AI Portraits

    Persona AI

    Studio headshots without the studio — the model learns your face from a handful of casual selfies, then delivers a finished gallery in the styles you pick.

  • AI Writing

    Quill

    Long-form AI writing that reads like you wrote it — define a voice profile once, and every draft comes back in your style instead of the model's default.

  • AI Assistant

    Streamline AI

    An AI assistant that never loses the thread — every conversation kept, models swapped mid-chat without losing context, and answers you can stop or rerun the moment they drift.

  • AI Workbench

    Helm

    An AI workbench for comparing models on one prompt: send it once and read the answers side by side, with personas that carry your tone. Your history lives in the browser, not on a server.

  • Conversational AI

    Amber

    One of our early GPT-powered builds — a Discord bot that learns a community's voice and keeps the conversation going when the server goes quiet.

How we approach it

The thinking done where it counts

The hard part of AI work isn't calling a model — it's deciding where the model belongs, grounding it in real data, and keeping a person in the loop.

  • Grounded, not guessing

    Retrieval over your data keeps answers tied to what's actually true for your business.

  • People stay in control

    Human-in-the-loop review and approval where the cost of a wrong answer is real.

  • Provider-agnostic

    We choose the right model per task and keep the integration clean, so you're never locked in.

Related services

Where this sits in what we do

AI is one focused strand of the work. Dig into the dedicated service, or see the full menu.

FAQ

Frequently asked questions

Yes. A lot of our AI work is integration — wiring a model into an existing product behind a clean API, with attention to cost, latency, and where the AI genuinely adds value rather than bolting on a chatbot.

Have a model-shaped problem?

Tell us what you're trying to do with AI. We'll be honest about where it helps — and build the part that does.

Start a project

Ready when you are

Tell us what you have in mind — a rough idea or a tight spec — and we'll tell you how we'd build it.