BEGIN:VCALENDAR VERSION:2.0 PRODID:-//HasGeek//NONSGML Funnel//EN DESCRIPTION:Hyderabad meet-up X-WR-CALDESC:Hyderabad meet-up NAME:In the Loop - harnesses\, loops\, graphs\, and the evals that keep th em honest X-WR-CALNAME:In the Loop - harnesses\, loops\, graphs\, and the evals that keep them honest REFRESH-INTERVAL;VALUE=DURATION:PT12H SUMMARY:In the Loop - harnesses\, loops\, graphs\, and the evals that keep them honest TIMEZONE-ID:Asia/Kolkata X-PUBLISHED-TTL:PT12H X-WR-TIMEZONE:Asia/Kolkata BEGIN:VEVENT SUMMARY:In the Loop - harnesses\, loops\, graphs\, and the evals that keep them honest DTSTART:20260905T050000Z DTEND:20260905T073000Z DTSTAMP:20260826T070311Z UID:session/37RQFkHr9ZH7fBs5CegHJb@hasgeek.com SEQUENCE:20 CREATED:20260802T124236Z DESCRIPTION:# Register for the meet-up\n**https://hasgeek.com/fifthelephan t/agent-harnesses-small-models-hyd-meetup/**\n\n---\n\n# Call for Proposal s: Agent harnesses & small models in production\nTwo years ago\, it was pr ompt engineering\, then context engineering\, then the harness: the tools\ , permissions and sub-agents you build around the model. Then the loop. No w the graph\, and state that has to survive a crash at step 40 of 60.\n\nE ach shift arrives with a wave of posts written by people selling a framewo rk\, and it takes about a year to find out which parts actually held. So: which of these abstractions are earning their keep\, and which are ceremon y?\n\nMeanwhile more teams are training and serving their own small models — and not all of them sit inside an agent. Sometimes it's to cut the bi ll\, often it's data that can't leave the building. Is a tuned 8B good eno ugh for your problem? How would you know?\n\nIf you've designed tools an a gent could actually use\, or wrestled a long-running job into surviving a crash — if you've fine-tuned a small model and measured what you gained and lost — if you've made inference cheap or fast enough to ship\, or bu ilt the evals that told you any of it was working — we want to hear from you.\n\n## Who should submit\nYou probably have a talk in you if you are: \n- building agent systems in production\n- working on harnesses\, evals\, or the infrastructure around them\n- fine-tuning or deploying small langu age models\n- optimizing inference — cost\, latency\, or both\n- measuri ng any of this from a product or engineering angle\n\nYou don't need to be an ML researcher. If you've shipped and operated something with models in the loop\, your experience is exactly what this audience wants.\n\n**Firs t time speaking?** This is especially for you. Editors review every submis sion and work with you to shape the talk — some of the best talks here s tarted as a rough idea.\n\n## What we're looking for\n- Talks grounded in hands-on experience\, not slides lifted from a launch post\n- Honest takes : what worked\, what didn't\, and what you'd do differently\n- Any domain: dev tooling\, healthcare\, legal\, financial services\, regional language s\, enterprise search — it doesn't matter\n\n**Format:** 30-minute talk + 10 minutes Q&A\n\n## Timeline\n- **Submission deadline:** 25 August 2026 \n- **Notification of selection:** 29 August 2026\n\n## How to submit\n**S ubmit your abstract** at https://hasgeek.com/fifthelephant/hyd-meetup-cfp/ sub.\n\nYour abstract should clearly describe:\n - the problem or topic being addressed\,\n - why it is relevant\,\n - who the session is inte nded for\, and\n - the key takeaways attendees can expect\n\n--- \n\n## About the editorial team\n**Chaitanya Sangani** - meet-up editor. [Chaitan ya is an independent AI consultant.](https://www.linkedin.com/in/chaitanya san/) He is a member of The Fifth Elephant community and has reviewed talk s for the Pune edition\, and at The Fifth Elephant 2026 Annual Conference. \n\n**Srix Sriramkumar** - reviewer. [Srix makes engineering teams AI-nati ve.](https://www.linkedin.com/in/srixsriramkumar/) Srix is a member of The Fifth Elephant community and has reviewed submissions for the [Enterprise AI in production meet-up](https://hasgeek.com/fifthelephant/enterprise-ai -in-production-meetup/videos)\, held in Bangalore in June.\n\n**Prasun Ana nd - reviewer** - [creator of Zasper\, a high-performance IDE for Jupyter Notebooks](https://www.linkedin.com/in/prasunanand/). Spoke at The Fifth E lephant 2025 Annual Conference on building a high-performance JupyterLab a lternative.\n\n**Madhusudhan Sambojhu** - reviewer. [Co-founder at Able & Uncommon](https://www.linkedin.com/in/madhusudhan-sambojhu-b9170022/)\; so ftware engineer working across web and mobile\, and an active reviewer in the Fifthel and Rootconf communities.\n\n**Raghotham Sripadraj - guest rev iewer** - [co-founder and CEO at Hypervelocity](https://www.linkedin.com/i n/raghothams/)\, leading work on small language models\, AI engineering\, and intelligent agents. Previously senior data scientist at Freshworks\, P ayPal\, and Ericsson\, and a past Fifthel editor.\n\n---\n\n## Contact\n- Email: info@hasgeek.com\n- WhatsApp / Phone: 7676332020 LAST-MODIFIED:20260826T015800Z LOCATION:Hyderabad - https://hasgeek.com/fifthelephant/hyd-meetup-cfp/ ORGANIZER;CN="The Fifth Elephant":MAILTO:no-reply@hasgeek.com URL:https://hasgeek.com/fifthelephant/hyd-meetup-cfp/ BEGIN:VALARM ACTION:display DESCRIPTION:In the Loop - harnesses\, loops\, graphs\, and the evals that keep them honest in 5 minutes TRIGGER:-PT5M END:VALARM END:VEVENT END:VCALENDAR