Adarsh Goel

Experience

Ola Group

Jan 2024 – Jan 2025

Program Manager, Founder's Office & BA Office — Bengaluru

Joined as Ola Group's youngest Founder's Office hire after graduating early from BITS Pilani; worked directly with Bhavish Aggarwal across Krutrim AI, Ola Electric, Ola Cell, and Ola Consumer.
Redesigned BA-facing group-level daily metrics, reducing metric packs to the highest-signal ~30% and driving report automation from 0% to ~90% for founder-review readiness.
Program-managed Ola's ONDC commerce deployment with BCG, including Varun Boppana, and internal product / ops teams, helping Ola Foods scale from 0 to ~6K orders/day and launch across 119 cities; coordinated 13+ ONDC grocery seller partners and tech-integration work to route food/grocery deliveries through Ola Consumer's ride-hailing and logistics supply.
Promoted to Program Lead for Ola's autonomous dark-store robotics pilot, a ₹60 Cr mandated New Commerce automation program using vision AI, robotic arms, grippers, conveyors, lifts, sensors, and warehouse-control systems, with ~₹10 Cr deployed across two built dark stores.
Drove the robotics pilot from founder-approved concept to working prototype in ~2 months for Ola Sankalp by managing a 70+ member cross-functional team across electrical, mechanical, software, automation, AI vision, robotics, procurement, manufacturing, finance, BCG, and business teams, coordinating senior leaders including Shaun Calvert, Perry Deu, and Ajit Gokhale.
Used cross-functional context to spot system-level efficiency opportunities; improved dark-store simulation throughput from ~720 to ~1,024 orders/day by correcting conveyor/sensor assumptions and modeling bidirectional conveyor movement.
Moved into BA Office program leadership for Ola Electric's Gen3, Roadster, and future-product portfolio, preparing BA/founder reviews across engineering, R&D, DFX, testing, drawing releases, homologation, warranty, and supplier decisions; unblocked an ABS SOP threat by switching to a new vendor, reducing time to SOP by 4 months.
Supported Ola Cell and battery-program workstreams with Vishal Chaturvedi across Pan-Asia equipment-supplier planning for Gigafactory, Battery Innovation Centre IPO-investor readiness, battery-warranty economics for new cells, and battery-fault RCA.

Glitch House

Jan 2026 – Apr 2026

Co-founder, Tech & Curriculum — Bengaluru

Co-founded an offline AI/robotics learning venture for middle-schoolers; owned product, curriculum, AI tooling, demo projects, workshop design, and technical storytelling around an AI-native project-based learning model, including a skill-tree curriculum and Gemini pipeline converting YouTube/Instagram project references into structured workshop manuals.
Validated demand through word-of-mouth workshops and demos, interacting with 120+ students and 40+ parents without paid marketing; parents organically referred other families after seeing projects like the AI Poetry Camera, shaping the model toward a subscription-based project studio and angel/fund conversations with Campus Fund, Alagu Periyannan, Phanindra Sama, and others.

AI Systems & Research

Independent AI Research: Agent Evals & World Models

Apr 2025 – Present

Self-directed exploration into model training for intelligent control and LLM failure-mode diagnosis through benchmark harnesses and eval improvements · Bengaluru

AutomationBench — agent-workflow evaluation and reliability. Deep-dived into Zapier’s real-world workflow-agent benchmark after frontier LLMs scored ~10%; identified brittle assertions and weak tool RAG as major bottlenecks.
- Built a multi-layer RAG pipeline with model-retrieval handshake, then added stronger system prompting and failure-aware prompt reinjection; improved same-model strict pass-rate by ~3× on a frozen 12-task slice.
- Converted the benchmark into an MCP server for subscription-based CLI agents including Claude Code, Codex, Gemini CLI, Opencode, and Openclaw, saving costs on independent eval runs compared to API pricing.
- Contributed fixes upstream to Zapier’s benchmark codebase; one PR merged, with other PRs currently being discussed with the authors.
Orbit Wars (Kaggle ML competition). Built a self-play RL bot for a strategy-based inter-agent game under severe Kaggle CPU limits. Offloaded deterministic game mechanics into a simulator tool inside the neural-network harness, so the model only had to learn strategic attack decisions; redesigned inference around a lightweight axial-attention transformer, cutting runtime from 3.16 s to ~20 ms per turn. Rule-based baseline ranked top ~450 globally; PPO self-play later reached 63% win-rate vs that baseline.
Modular world-models research framework (PyTorch). Inspired by Andrej Karpathy’s “autoresearch” framing, built a Hydra-based harness that constrains model-assisted research into explicit component contracts and targeted experiment configs: swappable encoders, representation learners, dynamics models, controllers, buffers, and workflows. Reproducing the world-model lineage from Ha & Schmidhuber’s 2018 World Models through Google DeepMind’s PlaNet and Dreamer series, with ablations over which model/scaffold combinations best support planning and control.

Leadership

Oasis — BITS Pilani's flagship cultural festival

Dec 2022 – Nov 2023

Festival Head (StuCCAn)

As one of 8 StuCCA festival heads, co-managed a ₹3.3 Cr student-run P&L for Oasis 2023, balancing artist bookings, publicity, festival inventory, campus visual identity, and 50+ department budget allocations; led the 60+ member Art, Design & Publicity department and closed the festival net profitable.

Conquest — India's first student-run startup accelerator (CEL, BITS Pilani)

Apr 2022 – Jan 2023

Web3 Track Lead

Conceptualized and launched a dedicated Web3 track; secured $20K in sponsorship; established partnerships with 5+ industry leaders; attracted 90+ startup registrations through organic marketing.

Education

Experience

AI Systems & Research

Internships

Awards

Leadership