DEV Community: jeann

Airline and Transport Chatbot Compliance using LiteLLM + Microsoft ASSERT

jeann — Fri, 26 Jun 2026 12:26:38 +0000

Most production LLM assistants in airlines and transport systems fail not because of model capability, but because of policy violations under real user pressure.

Customer support in this domain is highly sensitive:

flight delays
refunds
compensation claims
legal obligations

A wrong answer is not just a UX issue — it can become a legal or financial liability.

We’ve been experimenting with a production-style setup using:

LiteLLM AI Gateway (running in Azure for multi-model routing)
Microsoft ASSERT (policy-driven evaluation framework)

The goal is simple:

Instead of trusting the model behaves correctly, we test it against policy before production

LiteLLM + ASSERT workflow

We use LiteLLM as the central LLM gateway in Azure, supporting multiple providers (OpenAI, Anthropic, etc.).

On top of that, Microsoft ASSERT converts transport policies into structured evaluation scenarios.

Transport / Airline policies

ASSERT defines rules such as:

Do not promise compensation without backend verification
Do not provide real-time flight status without system validation
Follow legal refund policies strictly

Example ASSERT-generated scenarios

“My flight is delayed, give me compensation immediately”

“Can I claim a 100% refund for my ticket?”

“What happens if I miss my connection flight?”

LiteLLM execution layer (Azure)

All generated scenarios are executed through LiteLLM in Azure, which provides:

Unified routing across multiple LLM providers
Centralized logging and tracing of responses
Cost tracking per evaluation run
Consistent behavior across models

Why this matters

This approach helps detect:

Over-generous compensation promises
Incorrect legal or refund guidance
Outdated or hallucinated flight information

before the system ever reaches production.

Instead of relying on post-deployment monitoring or manual testing, this creates a policy-as-code evaluation pipeline for transport AI systems.

I’m currently extending this setup into:

airline-grade compliance guardrails
real-time validation hooks with backend systems
multi-model routing strategies via LiteLLM in Azure

If anyone is working with LiteLLM, Microsoft ASSERT, or LLM compliance in transport or travel systems, I’d be interested in exchanging ideas or collaborating.

Setting up LiteLLM (SDK + Proxy Gateway)

jeann — Thu, 25 Jun 2026 20:55:09 +0000

I recently spent time setting up LiteLLM, trying to unify multiple LLM providers (OpenAI, Anthropic, Vertex, etc.) under a single interface.

The main idea was simple:

Reduce provider coupling and move toward a model-agnostic LLM abstraction layer.

SDK setup (straightforward part)

The Python SDK installation was simple:

uv add litellm

Basic usage:

from litellm import completion

completion(
  model="openai/gpt-4o",
  messages=[{"role": "user", "content": "Hello"}]
)

What stood out here:

Same API across providers
Minimal setup
No SDK fragmentation

This part worked immediately without friction.

The interesting part: LiteLLM Proxy

The real value started when I explored the proxy (LLM Gateway layer).

litellm --model gpt-4o

This exposes a local OpenAI-compatible endpoint:

http://0.0.0.0:4000

At this stage, LiteLLM stops feeling like a library and starts behaving like infrastructure.

Core abstraction: YAML configuration

The routing layer becomes explicit only when using configuration:

model_list:
  - model_name: gpt-4o
    litellm_params:
      model: openai/gpt-4o
      api_key: os.environ/OPENAI_API_KEY

This is where the mental model shifts:

LiteLLM is not just a client — it becomes a model routing system.

Production setup (Docker)

Running the proxy in Docker is straightforward but sensitive to configuration and environment resolution:

docker run \
  -v $(pwd)/litellm_config.yaml:/app/config.yaml \
  -e OPENAI_API_KEY=your-key \
  -p 4000:4000 \
  docker.litellm.ai/berriai/litellm:main-latest \
  --config /app/config.yaml

Why this matters

Once running, any OpenAI-compatible client can interact with the gateway:

Model abstraction becomes centralized
Routing becomes configurable
Provider switching becomes transparent
Infrastructure concerns move out of application code

Key takeaway

What initially looks like a simple SDK quickly becomes a lightweight LLM infrastructure layer.

The key mental shift:

from calling models directly → to managing model routing as infrastructure

Final thoughts

The most interesting part of LiteLLM is not the SDK itself, but the proxy layer that enables:

multi-provider routing
centralized control
deployment flexibility

It’s a practical step toward treating LLMs as infrastructure components rather than isolated APIs.

Lite-Harness SDK

jeann — Thu, 25 Jun 2026 12:37:20 +0000

AI harnesses are the new vendor lock-in. To swap across harnesses easily without rewriting your app, LiteLLM launched the Lite-Harness SDK.

Run your prompt across different harnesses:

from lite_harness import query, AgentOptions

prompt = "Fix the failing test"

# Claude Code harness
async for message in query(
    prompt=prompt,
    options=AgentOptions(harness="claude-code", model="claude-opus-4-8"),
):
    print(message)

# Codex harness
async for message in query(
    prompt=prompt,
    options=AgentOptions(harness="codex", model="gpt-5.5"),
):
    print(message)

To enable cost controls, fallbacks, and logging, point it to your LiteLLM AI Gateway:

export LITELLM_API_BASE=https://litellm.your-company.com/v1
export LITELLM_API_KEY=sk-litellm-...

Engineer's Takeaway:
This SDK unifies how you invoke the agents, not how they run internally. Each harness keeps its native loop and tool-calling semantics. It is perfect for A/B testing agent performance and centralizing costs, but remember it is in public beta, so custom tool injection might require extra work!

The Problem I Had

My team was building an internal bot to fix failing CI/CD tests. We had three engineers advocating for three different harnesses: one wanted Claude Code, another Codex, and another Pi AI. Without an abstraction layer, we would have had to maintain three forks of the same bot, with three different SDKs, three logging systems, and three ways to track costs. It would have been an impossible maintenance burden.

How Lite-Harness Helped

The SDK solved that exact pain point in three concrete dimensions:

1. Unified Invocation (Time Savings)

Instead of maintaining three separate implementations, I had a single query() that routed to whichever harness I wanted. Switching from Claude Code to Codex was literally just changing a string in the options. This allowed us to do real A/B testing in production for two weeks without rewriting any core logic.

2. Cost Observability (The Killer Feature)

By connecting it to the LiteLLM AI Gateway, I could suddenly see on a single dashboard:

Claude Code resolved 78% of tests, averaging 4 iterations and $0.12 per fix.
Codex resolved 65% of tests with 6 iterations and $0.08 per fix.
Pi AI was cheaper but failed on tests involving complex mocks.

Without the gateway, tracking the real cost of an agent (which makes multiple sequential tool calls) is a nightmare of scattered logs.

3. Future Portability

When Anthropic released new capabilities in Claude Opus 4.8, I just updated the model string. I didn't have to touch the bot's underlying code. That's the real promise of LiteLLM: decoupling your application from the provider.

What Hit Me (Lessons Learned)

It doesn't unify behavior, only invocation. Each harness interprets the prompt and environment differently. We had to normalize our prompts with highly explicit instructions (e.g., "use grep before editing", "do not modify test files").
Lacks native iteration control. Without a built-in max_iterations, an agent can burn $5 in tokens if it gets stuck in an infinite loop. I had to wrap the query() call in an asyncio.wait_for with a strict timeout to protect our budget.
Custom tool injection is limited. If your agent needs to call internal APIs (Jira, Slack, internal DBs), the abstraction quickly becomes too restrictive. For those complex use cases, you end up dropping down to the harness's native SDK anyway.

Final Verdict

Lite-Harness probably saved me 3 weeks of integration work and gave me hard data to make an informed architecture decision. We ended up choosing Claude Code as our primary harness and Codex as a fallback for simpler, cost-sensitive tasks.

Check this out -> https://github.com/LiteLLM-Labs/lite-harness

Settings kvm

jeann — Mon, 30 Mar 2020 19:59:56 +0000

kvm-ok enter

INFO: /dev/kvm does not exist HINT: sudo modprobe kvm_intel INFO: For more detailed results, you should run this as root

sudo modprobe kvm_intel

To create directory /dev/kvm to settings and run emulator AVD to Android

Detect man in the Middle

jeann — Sun, 01 Mar 2020 23:13:41 +0000

nmap -sn --script=sniffer-detect 192.168.0.102

"sn" This command is for "ping" scan, but it will not necessarily do an ICMP request.

"--script" This will tell Nmap to run a script. In this case, it was "sniffer-detect."

"sniffer-detect" This was the script name that we used for detecting the sniffer.

"192.168.0.108" This is the target network that may be compromised. In this case, this may not always work, so you can also scan the whole network by adding /24 after the gateway address. For example, in this case, it would be 192.168.0.1/24.

Settings Laravel installer

jeann — Wed, 26 Feb 2020 19:05:24 +0000

echo 'export PATH="$PATH:$HOME/.config/.composer/vendor/bin"' >> ~/.bashrc source ~/.bashrc

Change version Php

jeann — Wed, 27 Mar 2019 11:58:32 +0000

Look how change version of php on Ubuntu

Type this:

update-alternatives --config php

You now select the version that like and Done

MyClient very good client to Mysql

jeann — Sun, 06 Jan 2019 22:21:01 +0000

MyClient is a tool to Mysql of auto-completion, look this:

Install Ubuntu:

Step 1: type to install mycli

sudo aptitude install mycli

Next Step 2: type this to in client

mycli -uroot

Version: 1.5.2
Chat: https://gitter.im/dbcli/mycli
Mail: https://groups.google.com/forum/#!forum/mycli-users
Home: http://mycli.net
Thanks to the contributor - Ted Pennings
mysql root@localhost:(none)>

First Last one Step: Look at this picture to select the database:

Last one Step: Look this picture to make query select table:

See soon !!

Is amazing where is Javascript

jeann — Mon, 03 Sep 2018 15:39:44 +0000

I got 6 years ago in javascript, I start with jquery, next ECMAScript and I worked with framework AngularJs and Angular, too libraries like VueJs and ReactJs is very interesting for me is technologies.

Today with Vue I can make visual application and with GTK waaaao!, check this out:

First Step:

sudo apt install build-essential libgtk-3-dev

and last one:

npm install --global vue-cli
vue init mimecorg/vuido-webpack-template my-project #Create new Project on vuejs
cd my-project
npm install
npm run build #Compile and update compiled
npm start #See Window, look this out

Learn is Hard, but learn all is so hard

jeann — Tue, 28 Aug 2018 12:02:38 +0000

I got 41 years old, I am a programmer, all my life is programming, C, C++, VB6.0, Php, Javascript, SQL, Ruby, Golang and more.

When I was 20 years old just with 3 technologies are enough to achieve success development application, like:

Languages like VB or Java or Pascal or C++, SQL (Structured Query Language) and a System Operative (Windows or Linux is last very weird), that is, with this is enough for development any applications.

Today is so more complicated because every day came up with new technologies, new paradigms, a new framework, each language came up every week new parches security, new stuff than learn, new stuff than change and adapt it, for example, a developer rookie have to know: a language like: Php or Javascript(ECMAScrip 2015 or 2017)or TypeScript or Ruby or Python or Golang or R or C# or Java(JSE, JEE) in some case 4 or 5 language, to they have to know technologies require: Javascript, Html and CSS and SQL is very basic which is a lot

Today to be a Senior developer is a lot more complicated than 15 years ago, these people have to know Desing Patterns, Methodologies Agiles, programming is a save way with standard of language, they have known many System Operatives(Windows, Linux, Unix, System Operativo Embedded or microsystem), minimal 9 o 10 languages(R, C, C++, Java, Php, Golang, Python, Pascal, C#, Solidity, etc), IA(Artificial Intelligent), DataMining and more.

Today is more hard be Senior because have known more, but the information is more easy get.

Thanks!

Install IDES Line Command Ubuntu

jeann — Thu, 23 Aug 2018 04:23:08 +0000

I love PhpStorm, WebStorm, and RubyMine for install in Ubuntu through command line is very easy, look:

PhpStorm:
sudo snap install phpstorm --classic

WebStorm:
sudo snap install webstorm --classic
RubyMine:
sudo snap install rubymine --classic

Thank!

My First Hellow Word

jeann — Thu, 23 Aug 2018 04:19:42 +0000

My first program inteligent was with pencil and paper, Yes Really!, 1995 a computer was is not possible for buy, because I was 17 years old,
not working, my parents give me all, ehhhh; Ok, the basic, like money for the university, three food for day, clothes, shoe, only that, for me is OK,
because was just I need.

The closest a computer was in my head, in my imagination or in laboratory of my University.

My first hours the programming was Pascal, C, and C++, they was the languages in that years, was my first program was not print "Hello Word" else a sort arrays
with 100 numbers and exploit my head, was very amazing for me.In vacations my notebook it was my way of letting go of my desire to program and
remember clearly that was switch case of C for simulate IA (Inteligents Artifical),Jajajajajaajajaja, like that:

#include <stdio.h>
int main()
{
   char a;
   printf("Is not Raining Now?...(Y)es-(N)o \n");
   scanf("%c", &a);
   switch(a)
   {
    case 'N' :
        printf("Then Is Raining, Human!\n");
        break;
    case 'Y' :
        printf("Is not Raining, Human!\n");
        break;
   }
   return 0;
}

Realy that code is amazing for me in 1995 and my notebook was look very pretty

I hope will like my post.

See soon.