Headline

Building an AI co-pilot · Omniful · Candidate Master, Codeforces (top 100 India) · 6★ CodeChef · LeetCode top 0.5% · 2× ICPC Regionalist · Building RiskKernel (open-source AI-agent reliability)

About

Currently building 𝗮𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜 𝘀𝘆𝘀𝘁𝗲𝗺𝘀, end-to-end. 𝗧𝗵𝗲 𝗦𝘂𝗻𝗿𝗶𝘇𝗲 — 𝐀𝐈 𝐜𝐨-𝐩𝐢𝐥𝐨𝐭 𝐟𝐨𝐫 𝐈𝐧𝐝𝐢𝐚𝐧 𝐫𝐞𝐭𝐚𝐢𝐥 𝐭𝐫𝐚𝐝𝐞𝐫𝐬. Six specialized AI agents — 𝘰𝘳𝘤𝘩𝘦𝘴𝘵𝘳𝘢𝘵𝘰𝘳, 𝘯𝘦𝘸𝘴, 𝘵𝘦𝘤𝘩𝘯𝘪𝘤𝘢𝘭𝘴, 𝘮𝘢𝘤𝘳𝘰, 𝘳𝘪𝘴𝘬, 𝘫𝘰𝘶𝘳𝘯𝘢𝘭 — debate every setup before a signal fires. A 𝗵𝗮𝗿𝗱-𝗰𝗼𝗱𝗲𝗱 𝗿𝗶𝘀𝗸 𝗲𝗻𝗴𝗶𝗻𝗲 enforces position discipline at the server. Trades execute via official broker APIs. 𝗜𝗻 𝗽𝗿𝗶𝘃𝗮𝘁𝗲 𝗯𝗲𝘁𝗮 𝘄𝗶𝘁𝗵 𝗶𝗻𝘃𝗶𝘁𝗲 𝗼𝗻𝗹𝘆. → 𝐡𝐭𝐭𝐩𝐬://𝐭𝐡𝐞𝐬𝐮𝐧𝐫𝐢𝐳𝐞.𝐜𝐨𝐦/ 𝗥𝗶𝘀𝗸𝗞𝗲𝗿𝗻𝗲𝗹 — an 𝐨𝐩𝐞𝐧-𝐬𝐨𝐮𝐫𝐜𝐞 𝐤𝐢𝐥𝐥 𝐬𝐰𝐢𝐭𝐜𝐡 + 𝐛𝐮𝐝𝐠𝐞𝐭𝐬 𝐟𝐨𝐫 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭𝐬. Deterministic 𝘤𝘰𝘴𝘵, 𝘭𝘰𝘰𝘱, 𝘵𝘪𝘮𝘦 budgets enforced in compiled 𝗚𝗼 — never by an LLM. Crash-resumable runs, human approval gates on side-effecting tool calls, OpenTelemetry GenAI export, BYO key. One env var to adopt. Apache-2.0. → 𝐠𝐢𝐭𝐡𝐮𝐛.𝐜𝐨𝐦/𝐩𝐫𝐚𝐬𝐡𝐚𝐫𝟑𝟐/𝐫𝐢𝐬𝐤𝐤𝐞𝐫𝐧𝐞𝐥 𝗕𝗼𝘁𝗵 𝘀𝗵𝗮𝗿𝗲 𝗮 𝗰𝗼𝗿𝗲 𝗶𝗱𝗲𝗮: deterministic risk engines wrapping non-deterministic systems. The safe part was never the smart part — it was the hard-coded layer around it. Before this I spent three years at 𝗢𝗺𝗻𝗶𝗳𝘂𝗹 — supply-chain SaaS, Gurgaon. Joined as the 𝗳𝗶𝗿𝘀𝘁 𝗯𝗮𝗰𝗸𝗲𝗻𝗱 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿 when the team was < 10. Stayed through it crossing 150. In that time I shipped: — 𝗢𝗿𝗱𝗲𝗿 𝗠𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁 𝗦𝘆𝘀𝘁𝗲𝗺 - processing 75K+ orders/day at 99.9999% uptime — 𝗦𝗵𝗶𝗽𝗺𝗲𝗻𝘁 𝗠𝗮𝗻𝗮𝗴𝗲𝗺𝗲𝗻𝘁 𝗦𝗲𝗿𝘃𝗶𝗰𝗲 - routing 500K+ shipments/month, internally called "the brain" — 𝗖𝗹𝗶𝗰𝗸𝗛𝗼𝘂𝘀𝗲-𝗯𝗮𝗰𝗸𝗲𝗱 𝗿𝗲𝗽𝗼𝗿𝘁𝗶𝗻𝗴 𝗮𝗻𝗱 𝗮𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀 𝗹𝗮𝘆𝗲𝗿 - answering 1M+ transactions/month — 𝗖𝗲𝗻𝘁𝗿𝗮𝗹𝗶𝘇𝗲𝗱 𝗟𝗼𝗴𝗴𝗶𝗻𝗴 𝗦𝗲𝗿𝘃𝗶𝗰𝗲 - every internal team adopted (cut incident response 40%) 𝗮𝗻𝗱 𝗺𝗮𝗻𝘆 𝗺𝗼𝗿𝗲... 𝗟𝗲𝗱 𝗺𝗼𝗻𝗼𝗹𝗶𝘁𝗵 → 𝗺𝗶𝗰𝗿𝗼𝘀𝗲𝗿𝘃𝗶𝗰𝗲𝘀 𝗺𝗶𝗴𝗿𝗮𝘁𝗶𝗼𝗻 end to end 𝗭𝗲𝗿𝗼 𝗱𝗼𝘄𝗻𝘁𝗶𝗺𝗲. 𝗭𝗲𝗿𝗼 𝗱𝗮𝘁𝗮 𝗹𝗼𝘀𝘀 I also do independent evaluation work on 𝗳𝗿𝗼𝗻𝘁𝗶𝗲𝗿 𝗟𝗟𝗠𝘀 — 𝘳𝘶𝘣𝘳𝘪𝘤 𝘥𝘦𝘴𝘪𝘨𝘯, 𝘳𝘦𝘴𝘱𝘰𝘯𝘴𝘦 𝘨𝘳𝘢𝘥𝘪𝘯𝘨, 𝘦𝘷𝘢𝘭 𝘱𝘪𝘱𝘦𝘭𝘪𝘯𝘦𝘴. The thread under all of it is 𝗰𝗼𝗺𝗽𝗲𝘁𝗶𝘁𝗶𝘃𝗲 𝗽𝗿𝗼𝗴𝗿𝗮𝗺𝗺𝗶𝗻𝗴. — Candidate Master on Codeforces (𝟮𝟬𝟰𝟴, 𝘁𝗼𝗽 𝟭𝟬𝟬 𝗶𝗻 𝗜𝗻𝗱𝗶𝗮). — 6★ on CodeChef. — Top 𝟬.𝟱% on LeetCode. — 𝗥𝗼𝘂𝗻𝗱 𝟮 at Google Code Jam, Meta Hacker Cup and 𝟮× ICPC Regionalist. — ~3,000 problems solved. Watching for 𝘁𝗲𝗮𝗺 — small, technical, building something I'd actually want to use and sell. adarsh.prashar32@gmail.com

Location

Bengaluru

AI stack

anthropic apis

claude code

claude

cursor

hermes agent

mcp

Work experience

Founding Software Engineer at Stealth AI Startup, Mar 2026 - Present · 6 mo, India, Building an 𝐀𝐈 𝐜𝐨-𝐩𝐢𝐥𝐨𝐭 𝐟𝐨𝐫 𝐈𝐧𝐝𝐢𝐚𝐧 𝐫𝐞𝐭𝐚𝐢𝐥 𝐭𝐫𝐚𝐝𝐞𝐫𝐬, solo and end-to-end. 𝗶𝗻 𝗽𝗿𝗶𝘃𝗮𝘁𝗲 𝗯𝗲𝘁𝗮, 𝗶𝗻𝘃𝗶𝘁𝗲 𝗼𝗻𝗹𝘆. — 𝗦𝗶𝘅-𝗮𝗴𝗲𝗻𝘁 𝗔𝗜 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲: 𝘰𝘳𝘤𝘩𝘦𝘴𝘵𝘳𝘢𝘵𝘰𝘳, 𝘯𝘦𝘸𝘴, 𝘵𝘦𝘤𝘩𝘯𝘪𝘤𝘢𝘭𝘴, 𝘮𝘢𝘤𝘳𝘰, 𝘳𝘪𝘴𝘬, 𝘢𝘶𝘵𝘰-𝘫𝘰𝘶𝘳𝘯𝘢𝘭. Multi-model debate (Claude + GPT) with per-user spend caps. — 𝗦𝗲𝗿𝘃𝗲𝗿-𝗲𝗻𝗳𝗼𝗿𝗰𝗲𝗱 𝗿𝗶𝘀𝗸 𝗲𝗻𝗴𝗶𝗻𝗲: 2% per trade, 6% monthly cap, 15% position size, 3 concurrent positions. Hard-coded, no overrides. — 𝗟𝗶𝘃𝗲 𝗯𝗿𝗼𝗸𝗲𝗿 𝗲𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻 via official Zerodha Kite API. Idempotent bracket orders, per-user serialization, paper/live toggle. — 𝗡𝗶𝗳𝘁𝘆 𝗔𝗧𝗠 𝘀𝘁𝗿𝗮𝗱𝗱𝗹𝗲 𝘀𝗰𝗮𝗹𝗽𝗲𝗿 with Black-Scholes pricing and India-VIX-derived IV for paper mode. 𝗦𝘁𝗮𝗰𝗸: Next.js 14, TypeScript, FastAPI, Python, Supabase, Razorpay, OpenAI, Anthropic. AES-256 for stored broker credentials. Full audit log of trades, risk checks, and LLM calls. → thesunrize.com ————————————————————————————————————————————— Also shipping 𝗥𝗶𝘀𝗸𝗞𝗲𝗿𝗻𝗲𝗹 — an 𝐨𝐩𝐞𝐧-𝐬𝐨𝐮𝐫𝐜𝐞 𝐤𝐢𝐥𝐥 𝐬𝐰𝐢𝐭𝐜𝐡 + 𝐛𝐮𝐝𝐠𝐞𝐭𝐬 𝐟𝐨𝐫 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭𝐬. — Deterministic 𝘤𝘰𝘴𝘵, 𝘭𝘰𝘰𝘱, 𝘵𝘪𝘮𝘦 budgets enforced in compiled Go — never by an LLM. — Crash-resumable runs, human approval gates on side-effecting tool calls, OpenTelemetry GenAI export, git-native + SQLite owned memory. — BYO key, no telemetry. One env var to adopt. Apache-2.0. 𝗦𝘁𝗮𝗰𝗸: Go core + Python SDK. Adapters for Claude Agent SDK, OpenAI Agents SDK. → github.com/prashar32/riskkernel ————————————————————————————————————————————— 𝗖𝗼𝗺𝗺𝗼𝗻 𝘁𝗵𝗲𝘀𝗶𝘀: deterministic risk engines wrapping non-deterministic systems.

AI agent rater and evaluator at Mercor, Feb 2026 - Present · 7 mo, Remote, Evaluation & Rubric Design: Design, author, and refine rubric-driven evaluation frameworks for AI agent and coding tasks, defining scoring criteria for reasoning, tool-use, and task completion used across frontier LLM training pipelines. Agent Evaluation: Rate AI agent outputs against these rubrics at scale, surfacing failure modes and improving inter-rater consistency across expert reviewer pools. Technical Specialist Work: Engaged as a Competitive Programming Specialist and SWE Expert, reviewing and grading advanced coding submissions for correctness, efficiency, and edge-case handling.

Software Engineer II (Backend + Data) at Omniful, Jul 2024 - Present · 2 yrs 2 mos, India, - Built the product-wide reporting & analytics layer on ClickHouse, delivering real-time insights for 𝟏𝐌+ transactions/month at sub-second 𝐩𝟗𝟓. - Designed and shipped the Shipment Management Service routing 𝟓𝟎𝟎𝐊+ monthly shipments — internally called "𝐭𝐡𝐞 𝐛𝐫𝐚𝐢𝐧" — improving delivery accuracy 𝟑𝟓% and automating complex multi-carrier logistics. - Built the in-house Centralized Logging Service adopted by every internal team; cut incident response 𝟒𝟎% and improved system observability at scale. - Led the monolith → microservices migration along with the underlying DB migration — zero data loss, zero downtime. - Drove platform stability and developer experience: SLO-driven monitoring, automated deployment workflows, migration playbooks the next 𝟏𝟎𝟎+ engineers joined into. - Operated as a senior IC, building platform tooling end-to-end as the engineering org scaled from 𝟏𝟎 𝐭𝐨 𝟏𝟓𝟎+ engineers.

Software Engineer at Omniful, Jun 2023 - Jun 2024 · 1 yr 1 mo, Gurugram, Haryana, India, - Built the Order Management System (OMS) processing 𝟕𝟓𝐤+ orders/day at 𝟗𝟗.𝟗𝟗𝟗𝟗% uptime — from first commit to high-volume launch. - Architected the Rule Management Service to handle dynamic business logic; cut feature rollout time 𝟓𝟎% and increased flexibility for 𝟖𝟎𝟎+ global sellers. - Built the Customer Management Service and Analytics Management Service on RocksDB, supporting 𝟓𝟎𝟎𝐊 active users with low-latency reads. - Led automated DB migrations, unified API versioning, and microservice scaling to handle a 𝟏𝟎× peak-traffic increase. - 𝐒𝐭𝐚𝐜𝐤: Go, RocksDB, Postgres, MongoDB, Redis, Kafka, AWS.

Software Engineer Intern at Omniful, Jan 2023 - May 2023 · 5 mos, Gurugram, Haryana, India, - Wrote the foundational Catalogue Management Service and Order Management Service — the backbone the commerce backend still runs on. - Built scalable catalogue ingestion and order processing pipelines serving 𝟖𝟎𝟎+ vendors and 𝟓𝟎,𝟎𝟎𝟎+ daily transactions. - Shipped high-quality Go microservices with Postgres / MongoDB / Kafka event-driven architectures, working alongside senior engineers from day one. - Gained hands-on IC experience shipping backend features from 𝟎 → 𝟏 and participating in the company’s earliest scaling phases

Problem setter at Scaler, Jun 2022 - Mar 2023 · 10 mos, Bengaluru, Karnataka, India, • Formulated over 400 problems for InterviewBit and Scaler Academy hiring contests, enhancing candidate evaluation. • Rigorously tested and validated 350+ problems, ensuring high quality and integrity in assessments. • Implemented optimal solutions for various problem sets, demonstrating strong algorithmic design and problem-solving skills.

Head Of Programming at SLIET Software Developement Club, Aug 2020 - Dec 2022 · 2 yrs 5 mos, Sangrur, Punjab, India

Education

Bachelor of Engineering - BE, Computer Science, Sant Longowal Institute of Engineering and Technology

Public links

github: https://github.com/prashar32

leetcode: https://leetcode.com/prashar32

codeforces: https://codeforces.com/profile/prashar32

youtube.com: https://youtube.com/@prashar32