Welcome to Issue N.06 of Jaseci Digest, a biweekly roundup of what's happening across the Jaseci and Jac open-source ecosystem.
|
|
|
| JacHacks A2Tech lands in Ann Arbor |
|
Jaseci Labs is hiring for the AI engineering team |
24-hour hackathon at the University of Michigan, inside the Leinweber Building. $10K+ in prizes, official hackathon partner of a2Tech360. |
|
Build production AI systems with the team behind Jac. Two roles open, hybrid in Ann Arbor and Detroit. |
| | September 26–27 this week · University of Michigan, Ann Arbor |
|
|
| | Hiring now • AI Software Engineer
• AI Software Engineer, Model Training |
|
|
|
|
|
| This week · Hackathon · Ann Arbor |
|
| JacHacks A2Tech lands in Ann Arbor |
24-hour hackathon at the University of Michigan, inside the Leinweber Building. $10K+ in prizes, official hackathon partner of a2Tech360. | | September 26–27 this week · University of Michigan, Ann Arbor |
|
|
JacHacks A2Tech lands in Ann Arbor
JacHacks A2Tech is a 24-hour, in-person hackathon at the University of Michigan, and Jaseci’s official hackathon partner for a2Tech360. It runs September 26–27, this week, with spots limited.
|
|
What to expect
|
🏆 $10K+ in prizes
|
|
⏰ 24 hrs build window, in person
|
| |
|
📍 UMich Leinweber Building, Ann Arbor
|
|
🤝 a2Tech360 official hackathon partner
|
| |
|
⚡ Limited spots, apply early
|
|
|
|
|
JacHacks’ fourth stop this year is a homecoming: back to the University of Michigan, where the very first JacHacks began. Bring a team, bring an idea, or come alone and find both there.
|
|
Jaseci Labs is expanding the team building production AI systems, and the Jac language they run on. Two roles are open right now, both full-time and hybrid between Ann Arbor and Detroit.
|
| Full-time · Hybrid · Ann Arbor / Detroit |
|
|
| Full-time · Hybrid · Ann Arbor / Detroit |
|
| AI Software Engineer |
|
AI Software Engineer, Model Training |
Build and ship the AI agents behind Jaseci’s production systems, and the knowledge bases they draw on. - Ship agents to production: tool calling, multi-step workflows, human-in-the-loop review.
- Build the enterprise knowledge base: ingestion and retrieval pipelines across dispersed data sources.
- Looking for: hands-on agent and retrieval experience, strong Python. Heavy use of coding agents is a plus.
|
|
Fine-tune and post-train the models Jaseci’s production systems depend on, then ship the agents built on them. - Fine-tune and post-train: SFT, LoRA, or DPO-style, evaluated against a baseline and deployed.
- Ship agents to production, embedded with customer engineering teams at Fortune 500 and startup scale.
- Looking for: hands-on fine-tuning and agent experience, strong Python. Serving infra (vLLM, TGI) or Kubernetes is a plus.
|
Full job description (PDF) → |
|
Full job description (PDF) → |
|
|
|
|
| Full-time · Hybrid · Ann Arbor / Detroit |
|
| AI Software Engineer |
Build and ship the AI agents behind Jaseci’s production systems, and the knowledge bases they draw on. - Ship agents to production: tool calling, multi-step workflows, human-in-the-loop review.
- Build the enterprise knowledge base: ingestion and retrieval pipelines across dispersed data sources.
- Looking for: hands-on agent and retrieval experience, strong Python. Heavy use of coding agents is a plus.
|
Full job description (PDF) → |
|
| Full-time · Hybrid · Ann Arbor / Detroit |
|
| AI Software Engineer, Model Training |
Fine-tune and post-train the models Jaseci’s production systems depend on, then ship the agents built on them. - Fine-tune and post-train: SFT, LoRA, or DPO-style, evaluated against a baseline and deployed.
- Ship agents to production, embedded with customer engineering teams at Fortune 500 and startup scale.
- Looking for: hands-on fine-tuning and agent experience, strong Python. Serving infra (vLLM, TGI) or Kubernetes is a plus.
|
Full job description (PDF) → |
|
|
Both roles: email [email protected] with your résumé and the role title as the subject line.
|
|
Pocketnest’s white-label financial wellness platform runs inside banks and credit unions. Building Birdie meant driving app-shaped APIs from open-ended conversation, in a context where nothing can be hallucinated and every action has to be traceable.
|
|
3 mo
concept to production assistant
|
|
6
agents under one orchestrator
|
|
4,579
members in a five-credit-union pilot
|
|
53%
rise in member wellness at MSUFCU
|
|
|
What they built
|
Six agents, one orchestrator
Typed Jac constructs, not config on a framework.
|
|
|
One language end to end
Middleware and by llm() agents, one type system.
|
|
|
Agent to agent, live
An MCP server and a JSON-RPC 2.0 A2A tier. MSUFCU’s own agent, Fran, calls Birdie directly.
|
|
|
Auditable by construction
Token passthrough keeps PII on its own side; a Recorder logs every step and tool call.
|
|
|
| |
“The experience was phenomenal. The unique ability to have the vision, have the collaboration, and being able to execute on that is unique as a partner.”
Chris Wascha, CTO, Pocketnest
|
|
|
8.5 hrs
build window, start to hard stop
|
|
77
projects submitted on Devpost
|
|
88%
had never written a line of Jac
|
|
73%
built with a coding agent
|
|
|
JacHacks SF ran on July 26 at Founders, Inc. The post-event survey is in, and the headline finding is how the code got written: 73% drove Jac through a coding agent rather than writing it by hand. All 77 projects are in the Devpost gallery.
|
|
How it landed
|
77% rated getting started 4 or 5 out of 5
|
|
69% felt more productive than their usual stack
|
| |
|
88% rated the event 4 or 5 out of 5
|
|
58% said yes to joining the Jac community
|
|
|
What builders liked most about the Jac stack
Post-event survey, 26 respondents. Multiple selections allowed.
| One language, full stack |
|
85% |
| Graph-based programming |
|
54% |
| Simplified database ops |
|
23% |
| Built-in authentication |
|
19% |
|
|
|
In the builders’ own words
| |
“It was really seamless, the agent understood the language and architecture much easily and wrote better more streamlined code overall.”
Survey respondent
|
| |
“I liked how I could do everything in one language and didn't have to move between frameworks.”
Survey respondent
|
| |
“The unified programming paradigm and the graph-based model make structuring full-stack AI workflows cleaner and more intuitive.”
Respondent with more than ten years of experience
|
|
 |
|
JacHacks SF Recap: 200+ builders, 77 projects in one day
A 90-second cut of the day in San Francisco: the room, the build window, and the 77 projects that came out of it.
|
|
 |
|
One Language, Every Stack: Jason Mars at JacHacks SF
A clip from the JacHacks SF stage: why one language covering the whole stack is what made same-day builds realistic for a room new to Jac.
|
|
|
| jaclang 0.37.19 | Latest release, breaking changes since 0.35 |
|
jac start and jac dev are gone. Serve with jac run --serve, dev with jac run --dev, deploy with jac scale deploy.- Type checking runs on every compile. No more
--no_typecheck escape hatch. jac test replaces pytest outright. Only .jac files are collected now.- Fixed-width numerics need explicit casts, e.g.
i32(n) into a sized parameter. [scale.microservices] is deleted. Workspaces ([apps.<name>]) replace it.
|
| Full release notes → |
|
|
|
|
| Join the Jaseci Discord |
|
Subscribe to Jaseci on YouTube |
Where the upgrade questions are getting answered right now. Debugging help, release chatter, A2Tech team-forming, and a steady stream of show-and-tells. |
|
Talks, demos, and the Jac tutorial series land here first. Subscribe to catch each new lesson as it drops. |
|
|
|
|
|
|
|
|
| Join the Jaseci Discord |
Where the upgrade questions are getting answered right now. Debugging help, release chatter, A2Tech team-forming, and a steady stream of show-and-tells. |
|
|
|
That’s Issue N.06. See you in Ann Arbor on September 26, and if one of those roles sounds like you, we’d love to hear from you.
|
|
|
Issue N.06 · September 23, 2026
|
| © 2026 Jaseci Digest. Part of the Jaseci open-source ecosystem. |
|
|