Sippin' lattes and shipping code

Unnati is building
Tessact AI & Polynomial.

Product engineer building AI tools across video, systems, and interfaces. At Tessact, that has meant generative AI pipelines, browser-based editors, and systems that turn complex AI workflows into usable products. Also built a content repurposing pipeline that turns long-form news into short-form, ready-to-publish video.

Get in touch →Bombay → Bengaluru

Now

Tessact AI — Video Intelligence

Software Engineer 2

Current Focus

Turning AI generated images and videos into marketing assets and talking avatars, and building the pipelines that make it all work.

View case studies →

Elsewhere

LinkedIn GitHub Twitter Medium

01Work

Shipping generative video and editors end to end.

Four years building here: React/TypeScript product surfaces (including a Remotion editor) and the AI pipelines behind generative video — avatars, script-to-video, repurposing, and production rendering.

Selected things shipped3 of 8

2026
Generative AI talking avatars & marketing assetsMulti-model orchestration (ElevenLabs, Seedance, Veo, Kling, NanoBanana) for lip-sync, voice cloning, music, pose-driven rendering, B-roll, and story-to-video — plus pipelines for image bundles, ads, and UGC-style creatives.
2025
Script-to-video & content repurposingEnd-to-end script-to-video with synthesis and automated music layering. Repurposing pipeline using transcription, vision LLMs, shot boundaries, and face tracking for short-form output with auto B-roll and metadata — cutting manual editing effort by about 85%.
2024
Browser video editor & Premiere Pro pluginReact, TypeScript, and Remotion editor with multi-track timelines, transitions, trimming, auto-captions, clipping, and multi-aspect reframing. Bolt CEP + React plugin for Premiere with bi-directional sync, comment-to-timeline mapping, and semantic clip search.

02Case studies

●2026·Gen AI · Pipeline architecture

Generative talking avatars, end to end.

A multi stage avatar generation pipeline behind a dedicated Gen AI service. Voice cloning, storyboard planning, background generation, lip sync, and stitching, coordinated across providers with resumable execution and stage by stage human review.

Pipeline stages: 11
Job types: 7
Provider adapters: 6

Read case study →ai avatars.unnati.studio

●2026·Gen AI · Brand & product systems

Marketing creative in four formats, one identity.

An orchestration layer that turns a product brief into image bundles, fast paced video, script to video, or UGC style ads. A product fidelity loop keeps the brand and product on model across every output mode.

Output modes: 4
Creation steps: 6
Capability adapters: 8

Read case study →ai marketing.unnati.studio

●2024·Gen AI · Video pipeline

Long form news into short form vertical clips.

An AI assisted clipping pipeline for newsrooms. Multi pass transcript analysis finds the story, computer vision keeps the right subject in a 9:16 frame, and producers stay in the loop for metadata, style, and final render.

Pipeline stages: 15
Analysis passes: 5
Output formats: 4

Read case study →clip engine.unnati.studio

source · briefing.mp447:22

00:00cuts · 6 detected47:22

Generated · 4 of 7

0:32

Quote — pricing

0:28

Stat — Q3 lift

0:31

Reaction — CFO

0:24

Closing line

06Product Videos

Assets

Rich media library management.

Media Pipeline

End-to-end ingest and transcode.

Remix

Real-time collaborative video editing.

Search v2

Semantic search across media assets.

Search

Fast full-text media discovery.

Tables

Sortable, filterable asset tables.

04Projects

Next.js · Drizzle · RAG · Langchain · OpenAI2025

Memoria

AI flashcard and MCQ generator for YouTube, PDFs, and PPTX extracts structure, builds context-aware decks. ~70 people using it.

Visit →

filter5 cols+ group

· · ·

task↕

owner

Shadcn data grid

Published React data grid: column pinning, drag-and-drop, resize, sort, and rich types (ratings, toggles, multi-select). Can handle 1000s of rows and columns.

Visit →

03About

AI keeps redrawing the map every few months. I'd rather be the engineer who keeps learning to read each new one than the one who memorised the old.

I'm a frontend engineer turned full-stack, working at the intersection of video and generative AI. Most of what I build helps marketing agencies and brands do more with the long-form content they already have.

Three years in at Tessact. Started on the frontend, picked up Python and Django, and now work mostly on content repurposing and the AI systems behind it.

Based: Mumbai → Bangalore
Working: SDE 2 · Tessact
Exp: 3.5 YOE
Focus: Video + Generative AI
Stack: TypeScript, React, Python, Django
Infra: Postgres, Redis, RabbitMQ, AWS
Reading: Shoe Dog by Phil Knight

On the side

Calligraphy

Copperplate on quiet evenings. Slowing down the hand to give the mind room.

Unnati Bamania

@asyncunnati · 2h

Introducing Tessact AI.

05Gallery

07Contact

Have a small, careful thing in mind?

Emailunnatibamania8@gmail.com

Bombay → Bengaluru · developing through 2026