akshat.devRésuméUmzobo ↗GitHub ↗
shipping umzobo
AI Engineer · Founder @ Umzobo

AkshatPortrait of Akshat GuptaGupta

✦ every face here is LoRA-generated

AI engineer and founder of Umzobo. I fine-tune LLMs, diffusion models and LoRAs, build RAG systems, and deploy generation pipelines on serverless GPUs.

See the work ↓
19projects
9certificates
8.63CGPA · Bennett
akshat — zsh — 80×20
next-token.json
akshat is a …
AI engineer0.72
founder0.19
filmmaker0.09
02 / about — cat README.md

Hi, I'm the human
behind the model.

~/akshat/README.md — Preview
# README.md

Akshat Gupta अक्षत गुप्ता

AI engineer working across generative AI, LLMs, VLMs, diffusion models, LoRA and computer vision. I fine-tune models, build RAG systems and ship production AI platforms — right now Umzobo, an AI media studio whose generation pipelines run on serverless GPUs.

Previously an AI Intern at LetsAI Solutions and a Generative AI Intern at Curlscape. B.Tech CSE (AI specialization) at Bennett University, CGPA 8.63. I like turning research into things people actually click on.

  • Trained a LoRA on my own face (every portrait here is generated)
  • Fine-tune custom Flux LoRAs for freelance clients
  • Shipped RAG chatbots over PDFs & vector DBs
  • Run video and image models as serverless GPU workers (ComfyUI on RunPod), some open-sourced
19projects shipped
4AI roles
9certificates
seed 4471 · lora v3
seed 4471 · lora v3
seed 0912
seed 0912
seed 2026 · cfg 3.5
seed 2026 · cfg 3.5
100% LoRA-generated ✦
03 / experience — git log --graph career

Commit history.

committed ✓
Founder @ Umzobo — AI Creative Studio
c6e5b82 · November 2025 → Present HEAD

Founder

@ Umzobo — AI Creative Studio

Built Umzobo end to end: an AI SaaS platform for media generation spanning video, image generation and editing, and creative workflows. Architected and deployed AI models and generation pipelines on serverless GPUs, with production auth, credits billing, custom LoRA support for client-facing use, and 50+ state-of-the-art models.

Generative AIServerless GPUsCustom LoRAFull-stackSaaS
focus
Generative AI
Serverless GPUs
Custom LoRA
Full-stack
SaaS
in progress
AI Intern @ LetsAI Solutions Pvt. Ltd.
2d418b4 · September 2025 → March 2026 merged

AI Intern

@ LetsAI Solutions Pvt. Ltd.

Helped develop, implement and fine-tune machine learning models, performed data analysis and preprocessing for training datasets, and worked with the team to optimise algorithms and improve model performance across AI/ML projects.

Machine LearningFine-tuningData PreprocessingModel Optimisation
focus
Machine Learning
Fine-tuning
Data Preprocessing
Model Optimisation
merged to main
Generative AI Intern @ Curlscape
cb92a94 · July 2025 → August 2025 merged

Generative AI Intern

@ Curlscape

Researched, evaluated and fine-tuned large language models (LLMs), and curated and prepared datasets.

Generative AILLMsFine-tuningEvaluationDatasets
focus
Generative AI
LLMs
Fine-tuning
Evaluation
Datasets
merged to main
committed ✓
Freelance LoRA Trainer @ Client work
5f361d5 · January 2025 → Present branch

Freelance LoRA Trainer

@ Client work

Fine-tune custom LoRA models for clients within Flux-based image generation systems, using Low-Rank Adaptation to deliver realistic, high-quality AI-generated images tailored to client needs.

LoRAFluxGenerative AIPython
focus
LoRA
Flux
Generative AI
Python
in progress
0000000 · 2022 — initial commit: started B.Tech CSE @ Bennett
04 / projects — 19 projects · scroll to browse

Things I've
shipped.

An IDE you scroll through. Every repo opens with a live preview, and the full case study is one click away.

akshat-gupta — ~/projects — Explorer
umzobo/README.md
demo.mp4
stack.toml
~/projects/umzobo/README.md

Umzobo

Founder · AI studio

An end-to-end AI SaaS for media generation: video, image generation and editing, and creative workflows. Generation pipelines run on serverless GPUs, with production auth, credits billing, custom LoRA support and 50+ state-of-the-art models.

Gen video & imageServerless GPUsCustom LoRATanStack · Bun · Postgres
Code private
await umzobo.directorsCut({ idea: "a heist in Mumbai" })
ai.umzobo.studio
Umzobo previewGPU Worker Fleet previewMiniMax H3 on RunPod previewLTX 2.3 Serverless previewComfyUI R2VA Nodes previewFine-Tuned LoRA Model previewDiffusion Training Studio previewMedical ChatBot previewCodeMind previewSmart Exam Assistant previewEye Disease Detection previewDistilling Qwen3 into GPT-2 previewTeacher-Student Distillation on IMDB previewFacePay previewAkshar previewKrishi Sanrakshan previewEnglish-Hindi Translation previewVision Transformer for MNIST previewCustom Emotion Generator preview
⎇ mainTypeScriptFounder · AI studio↑↓ browse · ↵ open
05 / skills — attention.visualize(skills, projects)

Where my attention goes.

attention_head_7.png — hover a token
UmzoboGPU FleetMiniMax H3LTX 2.3LoRA · FluxMedical ChatBotCodeMindSmart ExamEye DiseaseStory KDViT · MNISTEN→HI
skill × project · from stack + description
UmzoboGPU FleetMiniMax H3LTX 2.3LoRA · FluxMedical ChatBotCodeMindSmart ExamEye DiseaseStory KDViT · MNISTEN→HI
LoRA / QLoRA0.6·0.60.60.9·······
Diffusion0.6···0.9·······
Serverless GPUs0.90.90.60.6········
ComfyUI·0.60.60.6········
RAG·····0.90.60.9····
LangChain·····0.6······
LLMs·0.9···0.90.60.6·0.9··
Transformers··········0.90.9
Distillation·········0.6··
PyTorch····0.6····0.60.60.6
TensorFlow········0.6···
Computer Vision········0.9···
From each project's stack and description. Not a measured benchmark.
06 / academics — wandb run: btech-cse-bennett

Training run.

run/akshat-2022 · metric: cgpa
best 8.63final onlyrecorded final
8.63/ 10 · cumulative CGPA
programB.Tech CSE · Bennett Universityrun2022 → nowloggedfinal cumulative only · per-semester not entered
ckpt
now
B.Tech CSE · AI specializationBennett University · 2022–present
8.63
ckpt
2022
Class XII · CBSESant Gyaneshwar Sr. Sec. School, Delhi
80%
ckpt
2020
Class X · CBSEpre-training complete
95.2%
dataset: key_courses.jsonl
Advanced MLDL architecturesNLPComputer visionDSAProb & statsresearch: PEFT · RAG · multimodal
07 / certificates — open ~/Certificates

Proof of work.

Nine certifications, filed like a Finder folder. Click or tap one for Quick Look, then use ← → to flip through.

~/Certificates — 9 items
Certificates
08 / blog — ls ~/notes --sort=new

Field notes.

lora-on-my-face.mdx
Generative AI

I trained a LoRA on my own face. Here's every mistake.

Dataset curation, captioning, and rank/alpha sweeps on Flux — and why 18 photos beat 60.

noteread →
rag-pdf-chatbot.mdx
# retrieve → rerank → answer docs = loader.load("medical/*.pdf") db = FAISS.from_documents(docs) llm.invoke(prompt)
RAG

A medical RAG bot that knows when to say "I don't know"

Retrieve, rerank, and refuse. A local vector index over PDFs, with an answer only when the pages support it.

noteread →
vit-from-scratch.mdx
Computer vision

Vision Transformer from scratch on MNIST

Patches, heads, and a class token. A small ViT, written out, trained on digits.

noteread →
building-umzobo.mdx
Draft · soon

Building Umzobo: shipping an AI studio solo

Queues, GPU costs, and Bun on a small VPS. Notes from turning one idea into a film pipeline.

noteread →
09 / contact — open -a Mail

Let's ship something.

New Message
✉ Send⌘↵
To:Akshat Gupta <[email protected]>
From:
Subject:
POST /api/contact · sent to the inbox above
in
LinkedIn
↗gh
GitHub
↗
status
$ akshat --availability
→ open to: AI/ML roles
→ based in: India (IST, UTC+5:30)
✦ / ask akshat — RAG over résumé, projects & posts

Don't scroll.
Just ask.

A small RAG agent grounded only in this site. Every answer cites where it came from, and it says so when it doesn't know.

queryembedtop-k 4rerankstream
Ask Akshat
Ask about the work, the study, or how to reach Akshat. If it isn’t on this site, I’ll say so.
akshat.
© 2026 Akshat Gupta · built with Bun, Elysia & TanStack