Skip to main content
Back to Projects
Autonomous AI AgentsFeaturedFeb 2025 → Present

Autonomous eSIM Recommendation Agent

Telecom recommendation agent fine-tuned on Qwen2.5-3B with LoRA, quantized to GGUF via Unsloth for cheap, low-latency inference.

PythonQwen2.5-3BLoRAUnslothGGUFFastAPIPinecone
Model size
3B params
Quantization
GGUF Q4
Rec accuracy
+34% vs base

Problem & Solution

The Problem

Telecom customers need personalized eSIM plan recommendations, but rule-based recommenders miss context and cloud LLM APIs are too costly at scale.

The Solution

Fine-tuned Qwen2.5-3B with LoRA (Unsloth) on a curated telecom Q&A dataset, then exported to GGUF for quantized CPU/GPU inference. The agent runs as an autonomous service that returns ranked plan recommendations with explanations.

System Architecture

End-to-end flow from intake to outcome

  1. 01

    Dataset Curation

    Python

    Curated telecom plans, FAQs, and recommendation dialogues for fine-tuning.

  2. 02

    LoRA Fine-tune

    Qwen2.5-3BLoRAUnsloth

    Qwen2.5-3B base + LoRA adapters trained with Unsloth for fast convergence.

  3. 03

    GGUF Export

    GGUF

    Merged model exported to GGUF for quantized CPU/GPU inference.

  4. 04

    Retrieval

    Pinecone

    Plan metadata indexed for grounded, up-to-date recommendations.

  5. 05

    Inference Service

    FastAPI

    FastAPI exposes the autonomous recommendation agent with a typed contract.

Engineering Challenges & Wins

Optimization results and engineering tradeoffs

Optimization Results

  • LoRA fine-tuning with Unsloth for fast iteration on a 3B base model.
  • GGUF export for quantized, self-hosted inference without per-request API cost.
  • Autonomous recommendation loop with retrieval over plan metadata.

Engineering Challenges

  • Balancing fine-tune quality vs inference cost on a 3B parameter base.
  • Keeping the recommendation set fresh without retraining the model.

More case studies

Explore other production systems I've engineered end-to-end.

All Projects

Related Projects

Autonomous AI AgentsFeatured
OmniShield ICES: Asynchronous AI Email Security Middleware

Mail-server gateway (Postfix Milter) that intercepts every inbound message, runs a fine-tuned BERT core (M-BSCE) through an asynchronous inference pipeline (RabbitMQ + Celery), and adds multimodal threat detection (PaddleOCR for Quishing) with SOAP clawbacks to the upstream mail system.

Quishing detection
98%+
p95 classification
<400ms
Async throughput
5k msg/min
PythonPostfix MilterRabbitMQCeleryBERT+6
View Case Study
Autonomous AI AgentsFeatured
OpenClaw Multi-Channel Orchestration

Autonomous multi-channel AI platform that unifies chat, voice, and email behind microservices and an API gateway, with shared context across channels.

Channels
3 unified
Tenant onboard
<1 day
Cross-channel memory
100%
OpenClawFastAPIWebRTCWebSocketRabbitMQ+4
View Case Study