Building AI agents end-to-end · Voice AI · local LLMs · Data Pipelines · UX

Bangalore

Bangalore

Bengaluru

Bengaluru

Bengaluru







Bachelor of Business Administration - BBA, Business Administration and Management, General
bangalore university
Join Shayanto and 10k+ other professionals on
Tal is the app where bosses meet engineers & builders directly, without middlemen
Download Tal BossBuilding AI agents end-to-end · Voice AI · local LLMs · Data Pipelines · UX
I build AI agents end-to-end. Right now I'm in deep build mode — experimenting with the full stack of what a modern AI product actually takes to ship: → Custom agents (Claude, OpenAI, open-weight) wired with tools, memory, and real-world workflows via n8n and MCP. → Voice AI agents — Retell, Exotel, Ameyo, Fundamento — the messy intersection of telephony, ASR, LLMs, and barge-in interrupts. → Local LLMs — running open-weight models on my own hardware, fine-tuning them on domain data, comparing the cost/quality tradeoff vs frontier APIs. → Data pipelines that feed the above — ingestion, transformation, vector stores, eval loops. → Cloud architecture — designing the deploy layer so an agent isn't just a notebook, it's a production system. → Agent UX — the unsolved problem. How do humans actually interact with a system that can take 50 actions in a row? I'm prototyping interfaces that make that legible. The pattern across all of it: I learn by building the whole stack myself, then strip out what doesn't matter. Notebooks become repos, repos become deployed agents, deployed agents become writeups. What's interesting to me next: agents that own complete workflows in Indian fintech, voice-first products for users who don't read English, and the design language for AI surfaces that don't feel like chatbots. If you're building agents, training models, or designing AI products — let's talk. DMs open. Stack I work in: Claude, OpenAI, open-weight LLMs (Llama, Qwen, Mistral), n8n, MCP, Retell, Exotel, Python, LangGraph, Pinecone/pgvector, Docker, AWS/GCP, Figma, Next.js. Based in Bengaluru.
Bengaluru
claude code
claude
codex
cursor
mcp
n8n
Founder's Office at QodexAI, Aug 2026 - Present · 2 mo, Bangalore
AI Automation Builder at Independent, Feb 2026 - Aug 2026 · 6 mo, Bangalore, Built Scratchbook solo end to end, an AI life-OS live in production (27 RLS-private tables, 3 edge functions, Gemini backend). Built and operated autonomous systems solo, end to end: a 10-workflow AI "Chief of Staff, " voice outreach agents, two Alpaca trading bots, and a self-improving knowledge system. Built AI marketing automation end to end using Higgsfield mcp to automate marketing campaigns on Instagram.
Product Analyst at Ambak, Mar 2025 - Feb 2026 · 11 mo, Bengaluru, *Core Ownership & Scalability* End-to-End Funnel Ownership: Owned the analytics and operational logic for multi-channel communication funnels (Voice & WhatsApp), ensuring reliability and data consistency. High-Volume Operations: Supported 3 business verticals and 15+ active campaigns, managing thousands of monthly customer touchpoints for a team of 40-50 agents. System Scalability: Optimized system-level changes to improve throughput and operational predictability, allowing the business to scale without increasing headcount. *Product Design & Performance Optimization* Advanced Call-Flow Architecture: Designed and launched 10+ custom configurations, including agent-mapped DIDs and single-ACD routing. Connectivity Improvements: Boosted first-touch connectivity by 20-25% and reduced call misrouting by 30% through iterative system design. Lead Response Speed: Achieved a 30% reduction in time-to-first-contact across live campaigns. *Product Integration & UX Strategy* AI WhatsApp Integration: Partnered with Engineering to integrate Periskope (AI WhatsApp) into the CRM, contributing to iframe UX design and action placement. Efficiency Gains: Reduced agent context switching by ~40% by centralizing communication tools within a single interface. Stakeholder Collaboration: Translated agent-level behavior and funnel drop-off insights into actionable product hypotheses. *Growth & Customer Engagement* WhatsApp Funnel Engineering: Built automated engagement funnels using Meta-approved templates featuring CTAs and quick replies. Engagement Metrics: Drove 15-20% higher customer response rates through explicit state tracking. Automation: Reduced manual follow-up requirements by 25-30% through optimized messaging workflows.
Founder's Office at Nuttu Kaka, Oct 2023 - Feb 2025 · 1 yr 5 mos, Bengaluru, Founder's Office at Nuttu Kaka — a D2C consumer brand. The kind of role where you do every job once before it gets hired. • Organized 40+ product stalls at public events; engaged 300+ customers on product quality, health benefits, and pricing. • Managed a ₹25+ lakh corporate order for India Cements' Diwali gift boxes — owned sourcing, scheduling, packing, and logistics end-to-end. • Sourced products from 5 vendors across India; negotiated 2–5% cost reductions while holding quality and on-time delivery. • Ran competitor analysis on 15+ brands across Flipkart and Amazon — optimized pricing strategy and set 20–30% profit margins. Where a process didn't exist, I wrote it. The broad-surface experience here — product, ops, vendors, growth — is what shaped how I now think about wiring AI into the boring, repeatable parts of a business.
Analyst at Dashtoon, May 2023 - Aug 2023 · 4 mos, Bengaluru, Analyst intern at Dashtoon — a Generative AI content platform. My first hands-on exposure to building with generative models, before AI agents were the default conversation. • Tested and benchmarked Generative AI tools — Stable Diffusion 3, Midjourney, DALL·E 3 — for transforming and manipulating artwork inside a consumer-facing platform. • Consulted with the engineering team on UI design for the Stable Diffusion API, optimizing for prompt-to-render speed and creator efficiency. • Collaborated with the storyboard team to redesign a content workflow — increased team efficiency by ~30%. • Managed feedback from digital distributors; surfaced insights on content consumption across categories and demographics. • Contributed to the data analysis behind a regional heat-map feature that matched content to consumer preferences by geography and age group. Where the AI builder identity started.
Bachelor of Business Administration - BBA, Business Administration and Management, General, bangalore university
github: https://github.com/ShayantoDutta
shayanto-dutta.netlify.app: https://shayanto-dutta.netlify.app/