Skip to content
daksh.invoke()
role
AI Engineer
location
Jaipur, India
status
Open to new roles

Daksh Dhama

AI engineer. I ship LLM systems to production: RAG, multi-agent workflows and real-time voice pipelines.

shipped forButterCut AISeekhoReachall AIHitori Tech

/about

I'm Daksh, an AI engineer who likes the unglamorous half of AI products: the pipelines, queues, caches and evals that let a demo survive real traffic. I work in Python and TypeScript, mostly on LLM systems for early-stage startups.

I lead the backend at ButterCut AI, where I built the subtitle pipeline that turned our first enterprise pilot into a paying client. Before that I cut per-call LLM cost ~77% for production voice agents at Reachall AI. I care about cost per output, timestamps that line up, and agents that know when to say “inconclusive”.

797 contributions in the last year

/experience

3 roles

Backend Lead & Full Stack Engineer at ButterCut AI

Aug 2025 – Present

AI video post-production for short-form creators and enterprise teams.

  • First enterprise deal

    Closed the platform's first enterprise deal by shipping a production-ready video pipeline to Seekho, converting it into a paying client and validating the product commercially.

  • Two-pass STT architecture

    Reduced STT token usage by 89% with a two-pass architecture, cutting pipeline latency while preserving timestamp accuracy across multilingual audio.

  • Async B-roll pipeline

    Architected a five-stage async B-roll video pipeline (FastAPI, Docker) for parallel multimodal asset generation across sticker, text, poster and video overlays, backed by an AWS SQS worker with S3 offloading and containerized Lambdas at 99.9% uptime.

  • Retrieval for AI workflows

    Built RAG retrieval pipelines (embeddings, vector search) that ground LLM generation in retrieved context, integrated with the FastAPI services and PostgreSQL data layer.

Python / FastAPI / PostgreSQL / AWS SQS · Lambda · S3 / Docker / Deepgram / Gemini / Langfuse

Software Engineer (Part-time) at Reachall AI

May 2025 – Aug 2025

Production voice AI agents.

  • Post-call analysis cost ~77% lower

    Took per-call LLM analysis from $0.009 to $0.002 by removing duplicated prompt payloads, adding Gemini prefix/context caching and fixing pricing bugs across cached-token and audio-token tiers.

  • Sentence-level TTS cache

    Designed a response-caching layer (Redis/ElastiCache + S3, msgpack-serialized) that transparently wraps the TTS provider and eliminates redundant paid synthesis, rolled out platform-wide with per-agent toggles and cache-hit observability.

  • RAG for voice agents

    Built retrieval into production voice agents, using embeddings and vector search to ground LLM responses in retrieved knowledge.

Python / LiveKit / Gemini / Redis · ElastiCache / S3 / msgpack

Full Stack Engineering Intern at Hitori Tech

Dec 2024 – May 2025

  • Context-aware suggestions

    Integrated the OpenAI Chat API for context-aware suggestions, raising form completion by 35%.

  • Clearance Resume

    Developed backend systems and REST APIs for an AI-powered resume parser and builder, growing user sign-ups by 40%.

Node.js / Express / OpenAI / REST

Stack: Python, TypeScript, LangGraph, LangChain, FastAPI, PostgreSQL, PGVector, Redis, AWS Lambda, SQS, S3, Docker, LiveKit, Langfuse, OpenTelemetry, Node.js

Focus areas: multi-agent systems, retrieval (RAG), voice AI, cost engineering, evals, serverless pipelines, prompt & context caching

/education

B.Tech, Computer Science Engineering

NIMS University · Jaipur, Rajasthan

GPA 8.0 / 10 · Data Structures & Algorithms, Cloud Computing, System Design

2021 – 2025