Reliable agentic AI systems
VOLUME I · ONGOING
Anirudh
Sharma
Engineer, researcher, writer, maker — and several other things depending on the week.
A book about a person. The work is part of the story, not the title of it.
ANIRUDH SHARMAi
CONTENTS
The chapters so far.
Not a hierarchy. Not a final definition. Just the parts of life that have accumulated enough weight to deserve their own pages.
01EngineeringProduct systems · architecture · selected work12
02ResearchPapers · experiments · questions28
03LearningIITM · IIITB · LJMU · beyond38
04WritingEssays · notes · explanations46
05MakingDrawing · cinema · side projects54
06Stage & PeopleSpeaking · performance · community62
07AnirudhThe person between all of it70
TABLE OF CONTENTS1
FRONT MATTER · NOW
What has my attention right now.
How do we evaluate and diagnose tool-using agents under real failure modes?
First research paper published on IEEE Xplore — IEEE ICDH 2026
FRONT MATTER7
CHAPTER 01
Engineering
Build the thing. Then build the system around it.
1.1
AI/ML product engineering
My strongest professional thread is turning model ideas into capabilities that fit enterprise products, then making the surrounding patterns reusable for other engineers.
From model logic to product behavior
Built and evolved anomaly-detection capabilities inside enterprise billing workflows, spanning data preparation, model behavior, explainability, backend integration, performance investigation and stabilization after QA handoff.
Model engineering · SQL/data flows · performance analysis · release supportML that participates in a real product
Worked on predictive capabilities where training was only one layer: configuration, backend services, business behavior, model validation, governance documentation, environment support and product integration all mattered.
Predictive ML · backend integration · explainability/governance · issue triageStop rebuilding the same plumbing
Initiated and architected reusable ML foundations for data preparation, training, inference and configuration so multiple use cases and engineers could build on common patterns instead of bespoke pipelines.
Architecture · reuse · developer enablement · adoptionDesign it, integrate it, get it release-ready
Independently delivered AI capabilities across requirements, SQL/backend work, ML logic, application integration, engineering validation, deployment support, QA handoff and release readiness — while increasingly guiding freshers and interns.
Cross-stack ownership · technical direction · delivery through others1.2
Selected public work
Things you can inspect rather than take on faith.
CHAPTER 01 · ENGINEERING12
CHAPTER 02
Research
Does it actually work?
2.1
Published work
Research is where interesting ideas have to survive evidence.
Adverse Drug Event Prediction in Pharmacovigilance Using KG-Augmented LLM
My first published research paper: knowledge-graph-augmented reasoning for adverse drug event prediction using biomedical entity linking, semantic retrieval and structured evidence.
2.2
Experiments & current questions
Capability amplification, reasoning, evaluation and methods that improve AI systems without treating model scale as the only answer.
Inference-time strategies for code generation
HumanEval experiments across self-consistency, structured reasoning, planning, refinement, recursive criticism and test-based enhancement.
See experiments ↗Reliable agentic AI systems
Tool use, failure analysis, evaluation, retrieval, observability and the engineering required to make agents dependable in real products.
CHAPTER 02 · RESEARCH28
CHAPTER 03
Learning
The roads that taught me how to think.
3.1
Formal education
The degrees matter less as trophies than as the sequence they created: engineering foundations, formal AI/ML specialization, then research-oriented postgraduate work.
I
IIT Madras
B.Tech · Electrical Engineering
Minor in AI/ML
Minor in AI/ML
Engineering, mathematics, systems thinking and the foundations that came before the AI label became the center of gravity.
II
IIIT Bangalore
EPGP · Artificial Intelligence & Machine Learning
Formal specialization across machine learning and applied AI. Graduated as valedictorian.
III
Liverpool John Moores University
MSc · Artificial Intelligence & Machine Learning · Merit
Thesis: Prompt Engineering for Efficient Code Generation with LLMs.
3.2
What learning looks like now
Less collecting topics, more staying with difficult questions long enough to develop judgment: agents, evaluation, inference-time reasoning and reliable AI systems.
CHAPTER 03 · LEARNING38
CHAPTER 04
Writing
Ideas that escaped the notebook.
4.1
Essays & explanations
Technical essays, research notes and longer arguments. The website is the index; the writing can live wherever it was originally published.
The Harness Is The Product
Why the engineering around the model — retrieval, permissions, tooling, governance, observability and auditability — often determines whether enterprise AI actually works.
Read on LinkedIn ↗CHAPTER 04 · WRITING46
CHAPTER 05
Making
Not everything useful needs a KPI.
5.1
Things made for reasons other than work
Some things exist because making them changes how I notice, communicate or pay attention.
Drawing
Sketches, visual thinking and the pleasure of making something with no deployment pipeline attached.
Cinema
Indian cinema, visual storytelling, performance, memory, music and an unreasonable willingness to discuss all of it.
Side projects
Experiments that may begin as software, research, design, writing — or something I have not found a category for yet.
CHAPTER 05 · MAKING54
CHAPTER 06
Stage & People
Ideas change when another person is in the room.
Stage
Demos, talks, comedy, performance — I like the moment where an idea has to survive an audience.
Community
Gradient Ascent Syndicate, mentoring, teaching and building systems that let other people move faster too.
Explaining
Teaching and communication are not side effects of technical work for me. They are part of how I test whether I understand it.
CHAPTER 06 · STAGE & PEOPLE62
CHAPTER 07
Anirudh
The person between all the chapters.
Engineering is important to me. So are research, writing, making, cinema, stage, people and the occasional strange obsession. I like building structure around ambiguity, explaining hard things clearly, and making serious work feel a little more alive.
The best description is probably not a title. It is a pattern: learn something, build something, explain it, systematize it, and bring other people in.
engineeringresearchwritingdrawingcinemastagecommunity
“I do everything sincerely.
I take nothing seriously.”
CHAPTER 07 · ANIRUDH70
AFTERWORD
Good problem?
Good problem?
Interesting person?
If you want to talk engineering, research, agents, writing, open source, startup ideas, cinema, or something unexpectedly adjacent — say hello.
AFTERWORD76