Vardan Malik profile photo 4 on Tal
Vardan Malik profile photo 1 on Tal
Vardan Malik profile photo 2 on Tal
Vardan Malik profile photo 3 on Tal
Vardan Malik profile photo 4 on Tal
Vardan Malik profile photo 1 on Tal

Vardan Malik

Product & AI | Ex-DRDO, RRCAT, Kelly Technologies & AdGrid | Bennett University ’26 | Product · Analytics · AI/ML · Strategy

Delhi

< 1 year of work experience

Infinity Soft Systems
Sep 20262 moPresent

Noida

ADGRID
May 20263 moAug 2026

Gurgaon

Built "Chotu," an on-device Hindi/Hinglish conversational AI assistant enabling merchants to create discount and deal offers in natural language, running fully locally on a 3B LLM (Qwen 2.5 via Ollama) with ~1s response latency and no cloud dependency. Designed a "code for facts, model for reasoning" architecture — deterministic NLU parsers, value validation, and a fail-closed confirmation gate around the LLM — achieving 185/190 routing reliability and zero unsafe writes across multi-run regression suites. Selected the production model via head-to-head benchmarking (3B vs. 7B: 35/35 successful deal creations vs. freeze; ~1s vs. 10–17s latency), shipping the smaller model against the "bigger is better" default. Rejected two proposed architectures with measured evidence before building them: RAG (41% accuracy, 75% hallucination rate vs. curated retrieval) and an LLM intent-classification router (0/20 recall), replacing both with deterministic alternatives. Reverse-engineered a production REST API contract from Postman collections and internal docs, built a contract-accurate mock server, and migrated agent tool schemas to the real contract, verified via automated round-trip tests. Built Hinglish NLU components (deal-type detection, discount-trigger and duration parsing, input validation) with rigorous evaluation practice — multi-run verification, model pinning by digest, and recall reported on uncontaminated test sets (78%).
Shaip
Feb 20262 moApr 2026

Ahmedabad

Defence Research & Development Organisation (DRDO)
Jun 20252 mosJul 2025

Designed and implemented a bitstream-based image transmission system using Python to simulate serial communication between sender and receiver devices. Converted image data into byte-level streams and reconstructed images using NumPy, OpenCV, and pySerial, emulating embedded system communication protocols. Established virtual serial ports using socat to replicate hardware-level UART data transmission. Performed byte-wise data handling, metadata synchronization, and validation to ensure lossless image reconstruction. Gained hands-on exposure to low-level data transmission concepts relevant to embedded systems, surveillance, and defence communication systems.
Raja Ramanna Centre for Advanced Technology (RRCAT)
Dec 20241 moJan 2025

Indore, Madhya Pradesh, India

Built ML models to predict peak current in flash lamps from voltage inputs. Implemented Random Forest, Gradient Boosting, Polynomial Regression, and Neural Networks. Evaluated models to identify the most accurate predictor using industrial datasets.
Kelly Technology
Jun 20242 mosJul 2024

Analyzed user behavior data using SQL and Metabase, improving user retention by 8%. Designed the proprietary Kelly Score to quantify application engagement. Conducted usability testing and feature audits, increasing time spent by 17% and feature adoption by 4%. Built a structured FAQ and bug reporting workflow, reducing support tickets by 15%.

Vardan’s AI Stack

claude code
claude code
codex
codex
langchain
langchain
langgraph
langgraph
Figma Make
Figma Make
Qwen
Qwen

Education

Bennett University

Bachelor of Technology, Computer Engineering

Bennett University

20224 yrs2026

Join Vardan and 10k+ other professionals on

Tal Boss

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

Download Tal Boss

Vardan Malik on Tal

Headline

Product & AI | Ex-DRDO, RRCAT, Kelly Technologies & AdGrid | Bennett University ’26 | Product · Analytics · AI/ML · Strategy

About

AI Engineer building conversational AI systems. I’m a B.Tech CSE graduate from Bennett University, focused on Generative AI, LLM agents, RAG, NLP, and Voice AI. I enjoy building practical AI workflows using Python, LangChain, Ollama, APIs, prompt engineering, and evaluation-driven development. My focus is on making conversational systems reliable, grounded, and useful for real-world users. Currently looking for AI Engineer, GenAI Engineer, Voice AI Engineer, and AI Product roles. Core areas: Generative AI, LLM Agents, Conversational AI, RAG, NLP, Voice AI, Python, LangChain, FastAPI, Ollama, Prompt Engineering, LLM Evaluation.

Location

Delhi

AI stack

claude code

codex

langchain

langgraph

Figma Make

Qwen

Work experience

AI Engineer at Infinity Soft Systems, Sep 2026 - Present · 2 mo, Noida

Artificial Intelligence Intern at ADGRID, May 2026 - Aug 2026 · 3 mo, Gurgaon, Built "Chotu," an on-device Hindi/Hinglish conversational AI assistant enabling merchants to create discount and deal offers in natural language, running fully locally on a 3B LLM (Qwen 2.5 via Ollama) with ~1s response latency and no cloud dependency. Designed a "code for facts, model for reasoning" architecture — deterministic NLU parsers, value validation, and a fail-closed confirmation gate around the LLM — achieving 185/190 routing reliability and zero unsafe writes across multi-run regression suites. Selected the production model via head-to-head benchmarking (3B vs. 7B: 35/35 successful deal creations vs. freeze; ~1s vs. 10–17s latency), shipping the smaller model against the "bigger is better" default. Rejected two proposed architectures with measured evidence before building them: RAG (41% accuracy, 75% hallucination rate vs. curated retrieval) and an LLM intent-classification router (0/20 recall), replacing both with deterministic alternatives. Reverse-engineered a production REST API contract from Postman collections and internal docs, built a contract-accurate mock server, and migrated agent tool schemas to the real contract, verified via automated round-trip tests. Built Hinglish NLU components (deal-type detection, discount-trigger and duration parsing, input validation) with rigorous evaluation practice — multi-run verification, model pinning by digest, and recall reported on uncontaminated test sets (78%).

Product Management at Shaip, Feb 2026 - Apr 2026 · 2 mo, Ahmedabad

Machine Learning Intern at Defence Research & Development Organisation (DRDO), Jun 2025 - Jul 2025 · 2 mos, Designed and implemented a bitstream-based image transmission system using Python to simulate serial communication between sender and receiver devices. Converted image data into byte-level streams and reconstructed images using NumPy, OpenCV, and pySerial, emulating embedded system communication protocols. Established virtual serial ports using socat to replicate hardware-level UART data transmission. Performed byte-wise data handling, metadata synchronization, and validation to ensure lossless image reconstruction. Gained hands-on exposure to low-level data transmission concepts relevant to embedded systems, surveillance, and defence communication systems.

Machine Learning Intern at Raja Ramanna Centre for Advanced Technology (RRCAT), Dec 2024 - Jan 2025 · 1 mo, Indore, Madhya Pradesh, India, Built ML models to predict peak current in flash lamps from voltage inputs. Implemented Random Forest, Gradient Boosting, Polynomial Regression, and Neural Networks. Evaluated models to identify the most accurate predictor using industrial datasets.

Product Management Intern at Kelly Technology, Jun 2024 - Jul 2024 · 2 mos, Analyzed user behavior data using SQL and Metabase, improving user retention by 8%. Designed the proprietary Kelly Score to quantify application engagement. Conducted usability testing and feature audits, increasing time spent by 17% and feature adoption by 4%. Built a structured FAQ and bug reporting workflow, reducing support tickets by 15%.

Education

Bachelor of Technology, Computer Engineering, Bennett University