Public model-serving recipes, Windows installation reports and signed-bundle verification tests.
Project summary
I lead the Origo Labs team building Indra, an open-source AI workstation for Windows. This request funds a public release-validation effort: installation reports, reproducible local model-serving recipes, measurements and signed-model bundle tests. Those outputs would be available to anyone using or evaluating the code.
Indra already has public source, a Windows x64 1.8.0-rc.1 candidate, a handbook and automated tests. Independent installation testing and live serving with real model bundles remain unfinished. The desktop and document index run locally; models can be connected through hosted providers or compatible local inference servers.
Source: https://github.com/Artyfowl1710/indra_Origo-Labs
Handbook: https://artyfowl1710.github.io/indra_Origo-Labs/
Windows candidate: https://github.com/Artyfowl1710/indra_Origo-Labs/releases/tag/v1.8.0-rc.1
Goals and approach
At the $500 minimum, I would complete a smaller deployment test and public report: run a real small model bundle through the existing vLLM launcher, test altered-file rejection and document the exact environment, commands and outcomes. This is a proposed 2-4 week effort after funding and schedule agreement.
At the $2,000 goal, the proposed scope is 12 weeks: test installation with fresh profiles on two additional Windows PCs; exercise one llama.cpp setup and one Linux/WSL vLLM setup; fix reproducible failures and add regression tests; then publish scripts and raw synthetic-input results for timing, memory and successful-output throughput. The minimum project is contained within this larger scope.
How we will know it worked
The repository should contain a setup recipe with exact runtime, model and hardware versions, sanitized logs, stated pass/fail checks, and scripts another developer can run. The real-bundle test must distinguish an approved bundle from a modified one. Benchmark results will identify the workload and hardware, with failures and limitations recorded. This project does not establish universal model compatibility or a regulatory certification.
Funding use
Minimum $500: up to $250 temporary GPU/model-server test access, $150 additional Windows testing access and packaging checks, and $100 maintainer time for the deployment report.
Goal $2,000 total, including the minimum tier: $1,600 maintainer time, $250 GPU/model-server testing access and $150 Windows testing access and packaging checks. These are planning allowances rather than incurred costs or vendor quotes. Free or borrowed resources would be used first. The funded outputs are public code, documentation, test recipes and results. They would be released under appropriate open licenses with inherited notices preserved.
Team and track record
I am Mihir Khatri, age 20, founder and team lead at Origo Labs and the primary applicant. Indra is a team project; Origo Labs is unregistered, so I am applying as an individual for the stated maintainer work and test access. I study Computer Science and Engineering at SPIT Mumbai. My public CV includes a first-author Ro-SVD preprint on KV-cache compression, research internships at IIT Jodhpur and IIITDM Kancheepuram, and AI/ML industry experience.
CV: https://artyfowl1710.github.io/resume.pdf
Profile: https://github.com/Artyfowl1710
The MIT-licensed integration builds on earlier Indra_lok team work and Nous Research's Hermes agent framework. Original attribution is retained. AI tools have assisted development, documentation and this proposal.
Attribution: https://github.com/Artyfowl1710/indra_Origo-Labs/blob/main/NOTICE.md
Failure risks and outcomes
GPU or driver compatibility, package dependencies and machine access may prevent a documented setup from reproducing cleanly. The outcome could be a useful negative result and a narrower compatibility report. I would publish the failed reproducer and limitations, and agree any substantial scope or budget changes before spending further funds. Dates depend on funding and an agreed schedule. Energy measurements would be included only if suitable equipment is available and the measurement boundary can be stated.
Funding history and overlap
Indra has received $0 in project grants, prizes or sponsorships. A FOSS United request for INR 150,000 is submitted for overlapping release work. A $2,000 request has also been submitted to Lightcone Commons, with India payout constraints disclosed there. Other requests are being prepared. These are alternative or partial sources for one funding need: I would disclose awards and reduce overlapping costs or agree distinct milestones before accepting additional support.
FOSS United: https://fossunited.org/grants/directory/mihir_khatri
Full plan: https://github.com/Artyfowl1710/indra_Origo-Labs/blob/main/FUNDING_PLAN.txt
I live in India. The platform would need to confirm a lawful disbursement route and any charitable-purpose requirements before an award. My longer-term intention is a startup offering paid setup, deployment integration and support around the open-source core. This proposal funds the public release work, with the resulting code and evidence available to everyone.
Comments
No comments yet. Sign in to create one!