WeAreDevelopers Logo WeAreDevelopers
Sign in
For You LIVE
World Congress 2026 Europe • Jul 9, 2026 • Session details

Rules, Heuristics, or LLMs? Lessons from Solving the Same Problem Twice

Artur Naumenko

One engineering team built a drunk text generator twice. They pitted traditional heuristics against a fine-tuned LLM. Discover which approach truly minimizes hidden technical debt.

Pause
Mute Enter Fullscreen
#1 about 1 min

Solving the same problem twice with code and LLMs

Overengineering a project on purpose reveals the differences between rule-based logic and fine-tuned language models.

#2 about 5 min

Identifying the grey zone between heuristics and language models

Problems with subjective outcomes or infinite edge cases require choosing between probabilistic heuristics and data-driven models.

#3 about 2 min

Corrupting text as a test case for subjective metrics

Generating realistic typos provides a perfect test case because measuring text corruption relies purely on subjective observation.

#4 about 2 min

Using rule-based heuristics to simulate realistic typing errors

Combining Markov chains and edit distances creates a configurable but highly complex programmatic engine for text corruption.

#5 about 2 min

Training a language model with synthetically generated text pairs

Fine-tuning a lightweight model with LoRA requires thousands of examples generated by the initial heuristic engine.

#6 about 2 min

Evaluating output quality between algorithmic tuning and machine learning

While both methods produce believable results, rule-based systems often output extreme noise compared to the natural averaging of models.

#7 about 5 min

Balancing execution costs and maintenance flexibility in production systems

Algorithmic approaches offer cheap execution and deep debuggability while machine learning systems demand higher computational resources but handle complexity automatically.

#8 about 2 min

Four questions for deciding between code and language models

Analyzing output metrics, debugging requirements, input space width, and bounds definition determines the appropriate technical solution.

#9 about 2 min

Why handbuilt systems often require more effort than models

Hand-tuning complex probabilistic systems frequently proves more difficult than allowing a machine learning model to learn the desired behavior.

#10 about 3 min

Hardware requirements and tuning heuristics with machine learning models

Running lightweight models relies on consumer hardware while attempting to tune algorithmic parameters automatically introduces significant system fragility.

Matching moments

3:26 min

Introducing LLMs as judges for automated testing

Sebastian Messingfeld Sebastian Messingfeld · World Congress 2026 Europe

6:28 min

Natural language programming and prompt engineering

46 sec

Using LLMs to reverse engineer undocumented legacy code

Michele Zuccala Michele Zuccala +4 · World Congress 2026 Europe

1:53 min

Deciding if large language models are actually necessary

Mete Atamel Mete Atamel · World Congress 2024

1:59 min

Evaluating statistical models and the bitter lesson

Kamen Petroff Kamen Petroff · World Congress 2026 Europe

3:01 min

Balancing artificial intelligence tools with foundational software engineering skills

Tim Ruscica · Coffee With Developers

More from World Congress 2026 Europe

Building the Nervous System of AI - Michael Kagan (NVIDIA) thumbnail
29:17
about 2 months ago

Building the Nervous System of AI - Michael Kagan (NVIDIA)

Michael Kagan, Sead Ahmetović

Five Things in Tech that Matter Now - WeAreDevelopers World Congress 2026 Closing Keynote thumbnail
19:41
about 2 months ago

Five Things in Tech that Matter Now - WeAreDevelopers World Congress 2026 Closing Keynote

Christian Heilmann

The Sound of Your Secrets: Teaching Your Model to Spy, So You Can Learn to Defend thumbnail
32:06
about 2 months ago

The Sound of Your Secrets: Teaching Your Model to Spy, So You Can Learn to Defend

David vonThenen

Super scaling for the Super Bowl: How to survive 30 million users hitting your backend in 30 minutes thumbnail
35:13
about 2 months ago

Super scaling for the Super Bowl: How to survive 30 million users hitting your backend in 30 minutes

Irina Branovic

Running Secure Life Science Research at Scale using Hybrid GPU HPC and Kubernetes 🧬 thumbnail
24:23
about 2 months ago

Running Secure Life Science Research at Scale using Hybrid GPU HPC and Kubernetes 🧬

Jeremy Murray

Angstfreude - AI and Corporate Culture - The Thrill and the Threat thumbnail
30:26
about 2 months ago

Angstfreude - AI and Corporate Culture - The Thrill and the Threat

Axel Ebert, Karin Krobath

Upcoming sessions on this topic

Open session

World Congress 2026 North America

September 24, 2026 · 13:30–14:00

Stage 9

Understanding LLM Architectures: Inside the Design of Modern Models

Jofia Jose Prakash

Director - AI & Governance at Humanity + AI, Inc

Jofia Jose Prakash
Open session

World Congress 2026 North America

September 25, 2026 · 11:40–12:10

Stage 9

You Can’t Re-Run Sunlight: Designing ML Data Architectures for Physical AI

An Phan

Senior Data Infrastructure Engineer @ Hippo Harvest

An Phan
Open session

World Congress 2026 North America

September 24, 2026 · 16:10–16:40

Stage 9

Fast, Cheap, and Accurate: Optimizing LLM Inference with vLLM and Quantization

Legare Kerrison, Cedric Clyburn

Legare Kerrison
Cedric Clyburn
Open session

World Congress 2026 North America

September 24, 2026 · 17:30–18:00

Stage 6

No Single Model to Rule Them All: Building Resilient AI Agents Across Open & Closed LLMs

Emmanuel Acheampong

Senior Manager Developer Relations at Crusoe AI

Emmanuel Acheampong
Open session

World Congress 2026 North America

September 24, 2026 · 16:10–16:40

Outdoor Stage

When Humans Stop Writing Code: Rethinking Languages, Compilers, and Responsibility

Simon Auer

Organizer of flutter vienna meetup and CEO of marqably

Simon Auer
Open session

World Congress 2026 North America

September 25, 2026 · 14:10–14:40

Stage 4

Headroom: A Context Optimization Layer for LLM Applications

Tejas Chopra

Senior Software Engineer at Netflix

Tejas Chopra

Related videos

Stop Parsing Strings: Treating LLMs Like Type-Safe Microservices thumbnail
30:55
World Congress 2026 Europe • about 2 months ago

Stop Parsing Strings: Treating LLMs Like Type-Safe Microservices

André Behrens

Inside the Mind of an LLM thumbnail
27:11
World Congress 2025 • about 1 year ago

Inside the Mind of an LLM

Emanuele Fabbiani

Three years of putting LLMs into Software - Lessons learned thumbnail
26:30
World Congress 2025 • about 1 year ago

Three years of putting LLMs into Software - Lessons learned

Simon A.T. Jiménez

Cutting LLM Costs Without Cutting Quality: How to Beat Proprietary LLMs with Fine-Tuned Open Source thumbnail
29:50
World Congress 2026 Europe • about 2 months ago

Cutting LLM Costs Without Cutting Quality: How to Beat Proprietary LLMs with Fine-Tuned Open Source

Viktoria Semaan

Why LLMs Need Observability and How to Do It thumbnail
32:41
Europe 2026 Virtual • 2 months ago

Why LLMs Need Observability and How to Do It

Rahul Gupta

Command and Conquer: How we let an LLM control our Software thumbnail
37:51
Europe 2026 Virtual • 2 months ago

Command and Conquer: How we let an LLM control our Software

Simon A.T. Jiménez

Related articles

From learning to earning

Jobs that call for the skills explored in this talk.