Ferosh Jacob

Ferosh Jacob

Software engineering and AI leader focused on search, relevance, data science, and practical AI systems.

Posts

Part 4: E-commerce Alert: AI Shoppers Read Reviews, Not Your Store

5 minute read

Published:

Shoppers are starting to ask AI assistants what to buy, and agents that shop for them are next. I typed eight everyday product searches into Google, Bing and Claude on three evenings. The search engines’ links led to stores, Amazon above all. Claude built its answers from review sites, never opened an Amazon page, and named a store in fewer than a third of its answers. Where e-commerce stands today, and what changes when agents do the shopping.

Part 1: Why I Am Running These Experiments

5 minute read

Published:

AI assistants now answer ‘who should I hire?’ from indexes that are not Google’s, with lists that can change between runs. This series measures what that does to a business that paid for SEO.

Part 5: Scope Creep During the AI Age

4 minute read

Published:

AI agents made it cheap to start things, so the list of things in flight grows faster than one person can follow. Two projects of my own, 22 then 13 then around 100 open items, and what I’m taking from it: name the problem, do fewer projects, stay in the details.

Part 4: Providing Affordable SEO for Small Business Using Jev (Part 1)

8 minute read

Published:

The SEO tools that manage several sites at once price out a small business before it starts. So I’m building a semi-automated tool, with Jev narrowing ambiguous choices, that fills out business listings across ten sites without pretending every site allows a bot at the keyboard.

The Developer’s Job in Five Years: Conductor, Not Coder

10 minute read

Published:

Coding is getting cheap, but the developer’s job doesn’t vanish — it moves upstream into definition and downstream into validation. Three testable hypotheses for the next five years, including the counterexample that should make you skeptical of all three.

Software Is Getting Cheaper. But By How Much?

8 minute read

Published:

I’ve spent months arguing AI is making software cheaper. So I went looking for the evidence, applied my own rule, and found something more interesting than a percentage: software purchasing power.

CPU vs GPU battle, Round 1: leave the laptop, the ceiling comes with you

18 minute read

Published:

Three rounds of this experiment lived on a 16 GB laptop and nothing broke 0.5 QWK - a lexical floor at 0.222, and every learned or prompted arm crowded between 0.24 and 0.49. So we left the machine: the same prompt, the same parser, the same 3,000 held-out pairs, handed to a 7B judge on a datacenter GPU. It scored ESCI QWK 0.361 and WANDS 0.354 - better than the 3B it scaled up from, and still inside the same band. The ceiling is the task, not the model size. The result is now a poster at SCD 2026.

CPU vs GPU battle, Round 1: adapt the generalist, read the logits

16 minute read

Published:

Part 2 trained a specialist cross-encoder that won in-domain but leaned on its training data. This round we take the generative Llama-3.2-3B, LoRA-tune it on the laptop GPU with MLX, and grade by reading the label-token logits instead of prompting for JSON. In-domain it lands about even with the specialist (ESCI QWK 0.353), but it is the only arm that does not degrade out of domain - WANDS QWK 0.486, up from its own ESCI, while every other arm falls.

CPU vs GPU battle, Round 1: train the judge, don’t prompt it

8 minute read

Published:

In Part 1 we prompted a general model to grade e-commerce search relevance and watched it miscalibrate. This round we stop prompting and train a specialist - a bge-reranker cross-encoder fine-tuned on Amazon ESCI, on the laptop’s own Apple-Metal GPU. It becomes the first arm to beat the shipped on-device judge on both test sets (ESCI QWK 0.360, WANDS 0.299) at ~28 ms/pair in ~1 GB of RAM.

CPU vs GPU battle, Round 1: the agents brought the GPU

13 minute read

Published:

How well does a small prompted LLM grade e-commerce search relevance on a 0-3 scale, entirely on a laptop? We start the CPU-vs-GPU battle with the two easy arms - BM25 and a prompted local model - and find the honest surprise is not accuracy but calibration and circularity.

SLAB-RS, Part 8: The Word the Embeddings Couldn’t Read

9 minute read

Published:

Phase 3 measured the whole classical toolkit and shipped almost none of it. One thing did ship — and it didn’t come from a bigger model. It came from reading a thousand failures and noticing that a million-vector embedding model cannot read the word ‘without’. The fix was a single linguistic rule. +4.8 nDCG on negation queries, live.

SLAB-RS, Part 7: The Textbook Wins the Benchmark, and Loses the Trade

10 minute read

Published:

Retail is the first dataset in the project with real graded labels — so it is the first place the classical retail-search toolkit gets a fair audition. Field weighting stayed dormant. Learning-to-rank finally woke up, and then relearned the hybrid it was supposed to beat. The full textbook cascade won the benchmark — by +0.13% over the lean stack that was already shipped. Measure everything, ship almost nothing.

SLAB-RS, Part 6: The Agent Meets a Million Real Products

11 minute read

Published:

Five parts of a retail search series, and the documents were never products. Phase 3 changes that: 1.2 million real Amazon products, graded Exact/Substitute/Complement/Irrelevant labels, and one question — does the academic stack survive contact with real retail? It does. The BGE hybrid lands at parity with the published single-model baseline, at production latency, and int8 quantization makes a live million-vector index fit on a free-tier box, losslessly.

MS-SBN, Part 3: My Ten-Year-Old Built a Video Game From Four-Word Prompts

13 minute read

Published:

I typed two sentences into a coding agent, asked it to write a plan, and handed the laptop to my ten-year-old son. Over four weeks Mathai built a real, playable 2D game without writing a line of code. It was a deliberate experiment: proof that when every step has to verify it works, the software keeps working — whoever is driving. That is the core of Mission-Driven Engineering.

SLAB-RS, Part 5: How the Agents Discovered Hybrid Search

14 minute read

Published:

The agents took their Phase 1 wins to fifteen new domains — and discovered hybrid search: the one method that improved every single one, by up to 64%. The keyword tricks that looked brilliant on aeronautics turned out to be aeronautics-shaped, and even the cleverest ranker got demoted by Occam’s razor.

SLAB-RS, Interlude: Why the Agent Is Not in a Store Yet

16 minute read

Published:

Four articles into a series called retail search and there is not a single product — only aeronautics abstracts. Here is the map: what BEIR is, why the same BM25 scores 0.158 on one dataset and 0.789 on another, why I was wrong to call Phase 2 a filter, and what has to happen before the agent reaches real products.

SLAB-RS, Part 4: The Agent Breaks Its Ceiling with Embeddings

10 minute read

Published:

Embeddings from a chat model scored five times worse than random vectors. A purpose-built retrieval model beat the keyword ceiling by 8.4%. The agent’s three embedding attempts, a machine-learned ranking detour, and a live OpenSearch vector index.

SLAB-RS, Part 3: The Agent Climbs Keyword Search to Its Ceiling

11 minute read

Published:

An AI agent ran five ranking experiments on a live OpenSearch baseline in six days — work that would have taken a core search team months. This article shows the process, the failures, and how a rarely-shipped old technique won the keyword round.

MS-SBN, Part 2: Amy Turned a High-School Dream Into a Website

7 minute read

Published:

Amy Givens owned the MyAmy Designs domain since high school, but the website stayed a dream for years. With ChatGPT and Codex, she finally turned that domain into a real business website for MyAmy Designs & Remodeling.

Is Agile Failing in the Age of AI?

6 minute read

Published:

Agile transformed software development by adapting to changing requirements. But what happens when AI dramatically reduces the cost of implementation? Are we still optimizing for the right bottleneck?

The End of Software Scarcity, Part 2: Rosa’s Story

3 minute read

Published:

In my previous article, I argued that AI-assisted development may be bringing an end to software scarcity. This article tests that idea through a real-world project with a local small business owner.

Teaching Philosophy as a Part-Time Teacher

3 minute read

Published:

Between 2017 and 2025, I consistently taught undergraduate or graduate courses in most semesters while balancing my teaching responsibilities with a full-time position in industry.

Teaching philosophy

1 minute read

Published:

Teaching methods should be flexible, evolving based on the students, course content, and learning environment.