Abhishek Satish profile photo 1 on Tal

Abhishek Satish

Head of AI @ CarDekho Group · Agentic AI, LLM orchestration, production GenAI at scale

Jaipur

11+ years of work experience

CarDekho Group
Nov 20241 yr 10 mosPresent

Jaipur, Rajasthan, India

Built CarDekho Group's AI division from 0→1 post-acquisition. Grew the team 0→8 in ~18 months, mostly through internships. Set technical direction for agentic AI across CarDekho, BikeDekho, and enterprise clients in 5 countries. Production systems I built and own: • CarDekho cars chatbot (BikeDekho on the same stack) — user-facing AI assistant for India's largest auto marketplace. ~25K conversations/day, 1.5M+ lifetime messages, <3% hallucination, p50 latency 3–4s (down from 10–12s), blended cost-per-query ~₹0.4 vs ₹0.5 target. • Knowledge Engine — plug-and-play context system that predated MCP. Provides context, tooling, memory, RAG (vanilla → hierarchical → agentic), prompt management, structured outputs, external API connectors. FastAPI + Pydantic AI (migrated from LangChain/LangGraph). • Chat Intelligence — automated LLM eval and benchmarking platform. Nightly accuracy/hallucination/safety/context checks on sampled production traffic, surfaced via Grafana. Eval regressions gate prompt and model changes; every prod hallucination caught becomes an eval row. Guardrails (prompt-injection detection, content filtering, audit logging) layered on top. • OEM (American automotive) agentic data system — three-plane architecture: Data → Access (RBAC via Azure AD) → Agent, over an on-prem LLM stack inside the client's VPC. Wrote the PoC, won the contract, owned architecture, delivered MVP, handed off. • Other production work — sentiment summarizer API (500K+ reviews), document extraction (ViT + OCR + LLMs), news summarization, QLoRA fine-tunes on Gemma/Mistral for domain classification (+15–20% over base prompting). • Voice AI — PoC on Gemini Live + Twilio + ElevenLabs over WebSocket. Closed enterprise GenAI deals through hands-on PoC delivery. Rolled out the "Automind" platform across 5 countries. ~6 agentic/GenAI use cases shipped in 18 months — all non-internal use cases revenue-generating and profitable.
Toastmasters International
Jul 20161 yr 1 moJul 2017

India

Vicara
Apr 20168 yrs 8 mosNov 2024

Bengaluru, Karnataka, India

Founding CTO of Vicara, a deep-tech AI startup. Acquired by CarDekho Group in October 2024 — the acquisition was driven primarily by our LLM orchestration IP, agentic RAG, and the harness layer we used to ship to enterprise clients. What I built • LLM orchestration frameworks behind the acquisition: multi-agent orchestration (LangChain/LangGraph → Pydantic AI), tool-calling, ReACT agents with multi-step reasoning and error recovery, conversation state machines, RAG, memory, structured outputs, guardrails. • Large-scale distributed systems end-to-end: async processing, event-driven architectures, REST APIs, WebSocket streaming, PostgreSQL, AWS, Kubernetes. • Full-SDLC as founding technical leader — hardware, embedded firmware (ARM Cortex), backend, ML pipelines, cloud infra, agentic LLM frameworks. PyTorch + HuggingFace across time-series, anomaly detection, NLP, computer vision, sensor fusion. • Personally built every client PoC and converted them into multi-year enterprise contracts. Clients (Fortune 500) • Disney / Star Sports — real-time motion tracking for broadcast + content recommendation. • Maruti-Suzuki — sensor intelligence for manufacturing and customer engagement AI; strategic investor. • British American Tobacco — gesture-driven UX analytics and enterprise knowledge systems; strategic investor. • Ford — enterprise GenAI engagement. As a founder • Built the engineering team 0→15 across 7+ years and three product-market pivots. • Raised two funding rounds. Managed investors, board, accelerators (JioGenNext, Bosch DNA Nxt, SINE-IIT Bombay). • KAI wearable — launched on Indiegogo, raised $110K+ (212% funded), shipped to 35 countries. • 2 granted patents — US12229343B2 and GB2617708A (gesture / contactless digit recognition). • Y Combinator interviewed twice (S19, W20). Forbes India feature (Oct 2017). YourStory Tech 30.
Leaf Technologies Pvt. Ltd.
Dec 20152 mosJan 2016

Mumbai, Maharashtra, India

In my role as a Hardware Engineer at Leaf Technologies, I collaborated directly with the CTO on a pioneering project aimed at enhancing energy efficiency in HVAC systems. My responsibility was to develop a hardware product that could interface between HVAC controllers and the systems, optimizing energy consumption. Key Responsibilities and Achievements: Hardware Module Design: Led the design of the hardware module, tailored to improve the energy efficiency of HVAC systems. Firmware Development: Wrote high-performance firmware in embedded C for the STM32 microcontroller unit (MCU), ensuring reliable operation of the module. Connectivity Solutions: Integrated an ESP8266 module to provide seamless connectivity for the hardware. Algorithm Creation: Developed a bespoke algorithm that formed the core of the product's energy-saving capabilities. Data Recording and Simulation: Utilized MATLAB for recording data, visualizing, and simulating various scenarios to refine the product design. Schematic and PCB Design: Employed Altium Designer for meticulous schematic and PCB layout, crucial to the product's physical assembly. Project Highlight: Successfully accelerated the product's journey from concept to a Minimum Viable Product (MVP) within a tight two-month schedule. This project was a cornerstone in my professional development, significantly enhancing my skills in firmware programming, hardware design, thermodynamics, and various design and simulation tools. The experience was intense, challenging, and exceptionally rewarding, marking a key milestone in my career as a hardware engineer.
Toastmasters International
Jul 20156 mosDec 2015

India

Abhishek’s AI Stack

cursor
cursor
opencode
opencode
claudecode
claudecode

Education

Liverpool John Moores University

Master of Science - MS, Artificial Intelligence and Machine Learning

Liverpool John Moores University

20212 yrs2023
International Institute of Information Technology Bangalore

Postgraduate Diploma, Machine Learning & Artificial Intelligence

International Institute of Information Technology Bangalore

20201 yr2021

Join Abhishek and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Abhishek Satish on Tal

Headline

Head of AI @ CarDekho Group · Agentic AI, LLM orchestration, production GenAI at scale

About

Agentic AI engineer and leader — 9+ years in deep tech, 3+ shipping production LLM systems at scale. Today: Head of AI at CarDekho Group (India's largest auto marketplace, ~60M monthly uniques). Built the AI division from 0→8 over 18 months. Own the agentic stack that powers the CarDekho and BikeDekho chatbots (~25K conversations/day, 1.5M+ lifetime messages, <3% hallucination, blended cost-per-query ~₹0.4) plus an OEM agentic data system shipping into Fortune 500 automotive clients. Before that: Co-founder & CTO of Vicara (2017–2024) — a deep-tech AI startup acquired by CarDekho Group in October 2024 for our LLM orchestration IP. Built the team 0→15, raised two rounds, shipped to Disney/Star Sports, Maruti-Suzuki, BAT, and Ford. Two granted patents (US12229343B2, GB2617708A). Y Combinator interviewed twice (S19, W20). Forbes India feature, YourStory Tech 30. What I'm into right now: • Agentic system design — orchestration, tool calling, planning, RAG (vanilla → hierarchical → agentic), memory, structured outputs. • Production LLM engineering — Pydantic AI, FastAPI, evals as living artifacts, OpenTelemetry-based LLM tracing, cost/latency optimization at scale. • Agentic engineering as a practice — heavy daily use of Claude Code, Codex, and custom harnesses (ash, ash-v2) for spec-to-code, multi-file refactors, code review. Happy to talk shop on any of the above.

Location

Jaipur

AI stack

cursor

opencode

claudecode

Work experience

Senior Associate Director | Head of AI at CarDekho Group, Nov 2024 - Present · 1 yr 10 mos, Jaipur, Rajasthan, India, Built CarDekho Group's AI division from 0→1 post-acquisition. Grew the team 0→8 in ~18 months, mostly through internships. Set technical direction for agentic AI across CarDekho, BikeDekho, and enterprise clients in 5 countries. Production systems I built and own: • CarDekho cars chatbot (BikeDekho on the same stack) — user-facing AI assistant for India's largest auto marketplace. ~25K conversations/day, 1.5M+ lifetime messages, <3% hallucination, p50 latency 3–4s (down from 10–12s), blended cost-per-query ~₹0.4 vs ₹0.5 target. • Knowledge Engine — plug-and-play context system that predated MCP. Provides context, tooling, memory, RAG (vanilla → hierarchical → agentic), prompt management, structured outputs, external API connectors. FastAPI + Pydantic AI (migrated from LangChain/LangGraph). • Chat Intelligence — automated LLM eval and benchmarking platform. Nightly accuracy/hallucination/safety/context checks on sampled production traffic, surfaced via Grafana. Eval regressions gate prompt and model changes; every prod hallucination caught becomes an eval row. Guardrails (prompt-injection detection, content filtering, audit logging) layered on top. • OEM (American automotive) agentic data system — three-plane architecture: Data → Access (RBAC via Azure AD) → Agent, over an on-prem LLM stack inside the client's VPC. Wrote the PoC, won the contract, owned architecture, delivered MVP, handed off. • Other production work — sentiment summarizer API (500K+ reviews), document extraction (ViT + OCR + LLMs), news summarization, QLoRA fine-tunes on Gemma/Mistral for domain classification (+15–20% over base prompting). • Voice AI — PoC on Gemini Live + Twilio + ElevenLabs over WebSocket. Closed enterprise GenAI deals through hands-on PoC delivery. Rolled out the "Automind" platform across 5 countries. ~6 agentic/GenAI use cases shipped in 18 months — all non-internal use cases revenue-generating and profitable.

Assistant Director Of Public Relations at Toastmasters International, Jul 2016 - Jul 2017 · 1 yr 1 mo, India

Co-Founder, CPTO at Vicara, Apr 2016 - Nov 2024 · 8 yrs 8 mos, Bengaluru, Karnataka, India, Founding CTO of Vicara, a deep-tech AI startup. Acquired by CarDekho Group in October 2024 — the acquisition was driven primarily by our LLM orchestration IP, agentic RAG, and the harness layer we used to ship to enterprise clients. What I built • LLM orchestration frameworks behind the acquisition: multi-agent orchestration (LangChain/LangGraph → Pydantic AI), tool-calling, ReACT agents with multi-step reasoning and error recovery, conversation state machines, RAG, memory, structured outputs, guardrails. • Large-scale distributed systems end-to-end: async processing, event-driven architectures, REST APIs, WebSocket streaming, PostgreSQL, AWS, Kubernetes. • Full-SDLC as founding technical leader — hardware, embedded firmware (ARM Cortex), backend, ML pipelines, cloud infra, agentic LLM frameworks. PyTorch + HuggingFace across time-series, anomaly detection, NLP, computer vision, sensor fusion. • Personally built every client PoC and converted them into multi-year enterprise contracts. Clients (Fortune 500) • Disney / Star Sports — real-time motion tracking for broadcast + content recommendation. • Maruti-Suzuki — sensor intelligence for manufacturing and customer engagement AI; strategic investor. • British American Tobacco — gesture-driven UX analytics and enterprise knowledge systems; strategic investor. • Ford — enterprise GenAI engagement. As a founder • Built the engineering team 0→15 across 7+ years and three product-market pivots. • Raised two funding rounds. Managed investors, board, accelerators (JioGenNext, Bosch DNA Nxt, SINE-IIT Bombay). • KAI wearable — launched on Indiegogo, raised $110K+ (212% funded), shipped to 35 countries. • 2 granted patents — US12229343B2 and GB2617708A (gesture / contactless digit recognition). • Y Combinator interviewed twice (S19, W20). Forbes India feature (Oct 2017). YourStory Tech 30.

Intern at Leaf Technologies Pvt. Ltd., Dec 2015 - Jan 2016 · 2 mos, Mumbai, Maharashtra, India, In my role as a Hardware Engineer at Leaf Technologies, I collaborated directly with the CTO on a pioneering project aimed at enhancing energy efficiency in HVAC systems. My responsibility was to develop a hardware product that could interface between HVAC controllers and the systems, optimizing energy consumption. Key Responsibilities and Achievements: Hardware Module Design: Led the design of the hardware module, tailored to improve the energy efficiency of HVAC systems. Firmware Development: Wrote high-performance firmware in embedded C for the STM32 microcontroller unit (MCU), ensuring reliable operation of the module. Connectivity Solutions: Integrated an ESP8266 module to provide seamless connectivity for the hardware. Algorithm Creation: Developed a bespoke algorithm that formed the core of the product's energy-saving capabilities. Data Recording and Simulation: Utilized MATLAB for recording data, visualizing, and simulating various scenarios to refine the product design. Schematic and PCB Design: Employed Altium Designer for meticulous schematic and PCB layout, crucial to the product's physical assembly. Project Highlight: Successfully accelerated the product's journey from concept to a Minimum Viable Product (MVP) within a tight two-month schedule. This project was a cornerstone in my professional development, significantly enhancing my skills in firmware programming, hardware design, thermodynamics, and various design and simulation tools. The experience was intense, challenging, and exceptionally rewarding, marking a key milestone in my career as a hardware engineer.

Vice President Public Relations at Toastmasters International, Jul 2015 - Dec 2015 · 6 mos, India

Education

Master of Science - MS, Artificial Intelligence and Machine Learning, Liverpool John Moores University

Postgraduate Diploma, Machine Learning & Artificial Intelligence, International Institute of Information Technology Bangalore