I build things that think, see, and sometimes sell.
I'm Akshith Krishna — an engineer, founder, and product builder working across AI SaaS, computer vision, embedded systems, and human-machine interfaces.
Currently building AI tools, education platforms, trend engines, and systems that connect software intelligence with the real world.
Usually found shipping products, debugging C++, or convincing myself today is not leg day.
+ hardware
Built different. Literally.
My work sits at the intersection of AI, software, hardware, product, and human behaviour.
AI Products
Trend engines, education agents, voice coaches, RAG systems, and personal assistant workflows.
Computer Vision Systems
Industrial camera pipelines, embedded OpenCV, Jetson deployment, PTM/RTI imaging, and CUDA experiments.
Hardware + Human Interfaces
Power-over-skin research, RF circuits, wearable systems, PCB prototyping, and hardware-software integration.
Consumer Apps
Mobile apps, marketplaces, dating/social ideas, productivity tools, and experiments built from scratch.
Projects that made it out of the group chat.
Selected products, systems, and experiments across AI, computer vision, hardware, and startups.
TrendTune
AI-powered trend intelligence and automation platform for e-commerce brands. Tells you what to promote, restock, reposition, or launch.
Koras
AI education platform with personalised lesson plans, voice-based practice, and teacher-assistive workflows. Doesn't replace teachers — assists them.
Senswork Computer Vision System
Industrial computer vision: Baumer smart cameras, Jetson boards, OpenCV C++ pipelines, controlled lighting, and reflective transformation imaging.
Power-over-Skin (Final Year Project)
NTU FYP exploring wireless power transfer using the human body as part of the path. 40 MHz capacitive coupling, matching networks, and PCB iterations.
Some projects needed more than a landing page.
Deep dives into the messy technical parts: hardware, deployment, debugging, model evaluation, and real-world constraints.
Industrial computer vision on embedded hardware
Cross-compiled C++ pipelines running on Jetson, paired with a 64-light LED dome and Baumer cameras for surface analysis.
- Problem
- Industrial inspection on reflective parts: a single light + a generic camera can't capture enough information to detect defects reliably.
- What I built
- End-to-end embedded vision pipeline — Baumer neoAPI ingest, OpenCV + CUDA processing on Jetson Xavier NX, multi-light dome capture, PTM/RTI generation.
- Hardest challenge
- Cross-compilation for ARM64 Linux + keeping the LED dome capture sequence in lockstep with the camera shutter.
- What I learned
- Embedded vision lives or dies on lighting and timing. Algorithms come third.
Power-over-Skin RF system
40 MHz capacitive coupling through the human body, tuned with VNA-measured matching networks and a Schottky voltage-doubler rectifier.
- Problem
- Wearables and implantables need power without bulky batteries — the body itself can be part of the link.
- What I built
- TX/RX hardware, matching networks, PCB iterations, and a rectifier chain that lit an LED through the body.
- Hardest challenge
- The body channel changes with skin contact, hydration, posture. Matching is a moving target.
- What I learned
- RF intuition + a VNA beats simulation alone. Tight, fast iteration loops on real boards.
RAG vs base LLM evaluation
Built a fair evaluation harness for a RAG pipeline vs a base LM — same prompts, same compute envelope, BLEU / ROUGE-L / METEOR + latency.
- Problem
- Does retrieval actually help, or is it just vibes?
- What I built
- End-to-end RAG pipeline (embeddings → FAISS → context → LLM) and an evaluation harness measured against the base model.
- Hardest challenge
- Designing fair evaluation — same prompts, same generation params, same compute envelope.
- What I learned
- RAG wins where the base model lacks the facts. It costs you where the model already had them.
My stack changes. The shipping habit doesn't.
AI / ML
Models, agents, and pipelines.
Software
Web, mobile, backend.
Computer Vision / Embedded
Cameras, Jetson, low-level pipelines.
Hardware
RF, PCBs, wearables.
Product / Business
From idea to GTM.
Less resume. More lore.
I'm Akshith Krishna, a final-year Electrical & Electronic Engineering student at NTU. I've been building apps for over 8 years and have shipped more than 100 apps, experiments, tools, and product ideas.
My work sits somewhere between AI founder, computer vision engineer, hardware tinkerer, and product-obsessed builder. I like taking messy ideas and turning them into something people can actually use.
I've worked on everything from AI SaaS platforms and education tools to industrial camera systems, embedded C++ pipelines, RF hardware research, and consumer app concepts.
Outside of building, I'm probably at the gym, overthinking product positioning, or convincing myself that this is definitely the final version of the pitch deck.
Founder.
Engineer.
The build log.
Notes from building products, breaking things, fixing them, and pretending it was part of the plan.
How I built an AI trend engine for e-commerce
Trends are everywhere. Useful trends grounded in your actual catalog are not. Here's how TrendTune bridges that.
What industrial computer vision taught me about real-world software
When your pipeline runs on a Jetson under a 64-light dome, every lazy assumption shows up at 3am.
Building AI tools that are not just wrappers
The 'thin wrapper' meme is half right. Here's where the real moat lives in AI products today.
My Power-over-Skin FYP, explained simply
Wireless power transfer using your body as part of the path. No, it doesn't electrocute you.
Lessons from building 100+ apps
Most of them died. The pattern across the survivors surprised me.
How I think about startup ideas
A scrappy framework for filtering 'cool' from 'will actually be alive in 12 months'.
Building something interesting?
Reach out if you're building something ambitious in AI, commerce, education, computer vision, hardware, or human-machine interfaces.
Especially if it involves code, caffeine, weird hardware, or a product idea that sounds slightly unreasonable.