Round 2 applications are now open. Applications close 26 July 2026.
Where you'll find our alumni...
.png)
_edited.png)
What is TARA?
The Technical Alignment Research Accelerator (TARA) is a 14-week part-time program that builds technical AI safety skills for talent across the Asia-Pacific region (APAC). Twice a year (March-June and September-December), participants work through the ARENA curriculum via full-day Saturday sessions plus 2-7 hours of independent learning during the week.​
​
You'll learn and implement key concepts like transformer architectures, mechanistic interpretability and alignment research techniques, reinforcement learning, and model evaluations. Working in groups of 10-20 participants, you'll practice pair programming - where two people collaborate on code together, taking turns to write and review - all under the guidance of an experienced Teaching Assistant (TA)​.
​
The program is completely free—we cover:
-
Expert TAs
-
Saturday lunches
-
Compute and API credits for coursework and projects
-
Dedicated study spaces in your city
Round 1 2026 ran in Melbourne, Sydney, Brisbane, Manila, Tokyo and Singapore.
​
For Round 2 2026, we're targeting more cities across APAC. Where TARA runs depends on demand - if you want a TARA cohort to run in your city, let us know in the application form!
Cities we're considering if demand is high enough:
Melbourne, Sydney, Brisbane, Manila, Tokyo, Singapore, Taipei, Hong Kong, as well as cities in New Zealand and India.
Why apply to TARA?
-
Build deep technical understanding. You'll implement core ML and AI safety algorithms from scratch - not just use libraries, but understand how they actually work. This foundation is essential for research and engineering roles in the field.
-
Learn with expert support. Through structured Saturday sessions and ongoing Slack support, you'll receive guidance from an experienced TA on technical concepts and project development.
-
Produce real research. The program culminates in a three-week project where you'll receive detailed guidance from your TA and peers, with the opportunity to produce publication-quality work.
-
Join a regional network. We're aiming to bring together 80+ participants across APAC twice a year. Through small group sessions, collaborative projects, and ongoing discussions, you'll build lasting connections with others passionate about AI safety.
Participant outcomes so far
-
8 city cohorts
-
82 graduates
-
90% likelihood-to-recommend
-
95% very or extremely satisfied with program management
-
91% more likely to pursue a career in AI safety
​​
12-month career outcomes from 2025 pilot*
-
40% secured competitive fellowships (MATS, SPAR, LASR Labs, CSIRO Data61, Entrepreneurs First)
-
47% transitioned to AI safety roles (Australian AISI, ASET, Lyptus Research and EquiStamp)​
​
*15 of 19 graduates surveyed
For a full report on the TARA 2025 pilot program, follow this link.
"TARA gave me the technical foundation and community I needed to make the leap into AI safety. The structured curriculum, TA support and accountability of working alongside motivated peers made all the difference.
Within months of completing the program, I left my role at the Commonwealth Bank to join Arcadia Impact as an Inspect Evals maintainer."
— Scott Simmons, Arcadia Impact
Who is TARA for?
-
Software engineers and machine learning practitioners aiming to transition into AI safety.
-
Undergraduate and postgraduate students.
-
Technical professionals who can’t commit to full-time programs.
These are examples, not requirements. If you are a strong coder and passionate about AI safety, we encourage you to apply.​
Locations and dates
Round 1 2026: 7 March - 13 June
​
-
In Round 1 2026 we ran cohorts in Sydney, Melbourne, Brisbane, Manila, Tokyo and Singapore with ~80 participants across our six cities.
​
Round 2 2026: 5 September - 12 December
-
Applications are now open for Round 2 2026.
-
In addition to the cities from Round 1 2026, we'll also be targeting Taipei and Hong Kong, as well as cities in New Zealand and India. Final locations are determined by where our strongest applicants are based - if there's sufficient demand in your city, we'll make it happen.
-
We'll keep applicants updated throughout the selection process. Our goal is to make TARA accessible to talented participants across APAC while maintaining high-quality, in-person study groups.
​​
​
​​​
Curriculum overview (high-level structure)
The program follows a progressive learning journey through technical AI safety concepts, based on the ARENA curriculum. Topics for Round 1 included:
​
-
Neural network foundations and optimisation
-
Transformer architectures
-
Mechanistic interpretability
-
Reinforcement learning
-
Model evaluations
The final three weeks are dedicated to project work with TA support. A detailed weekly breakdown can be found here.
​​
The ARENA curriculum is dynamic, so our curriculum may change between rounds. TARA's Round 2 2026 curriculum is not yet finalised and may include new material from the most recent ARENA curriculum update.
Weekly learning structure
Each week follows a consistent pattern designed to maximise learning while accommodating participants' schedules. Start times vary by city and may shift slightly: 9:30am in India, 9:30am in Australia, 10:00am Singapore/Manila/Tokyo/Hong Kong/Taipei, and 10:30am in New Zealand.
Saturday Sessions (~ 7.5 hours):​​
-
Hour 1: TA introduces the week's technical concepts
-
Hours 2-3: Pair programming to implement learnings
-
Midday: Progress update, then lunch
-
Hours 4-6: More pair programming
-
Afternoon: TA presents on an AI safety topic
-
End of day: Progress update and wrap-up
Remote TA support is available on demand throughout the day.
Independent Learning (2-7 hours during the week):
-
Self-paced study of concepts introduced on the Saturday
-
Ongoing Slack support from TAs and fellow TARA participants
This structure ensures participants receive structured guidance while maintaining flexibility for individual learning styles and work commitments.
What are the entry requirements?
​Important:​
-
Strong English proficiency (all instruction and support is delivered in English).
-
Motivated to reduce risk from AI.
-
Strong Python programming background.
-
Attend at least 11 of 14 Saturday sessions in person (necessary for certification).
-
Arrive on time each Saturday (for Round 2 the times are 9:30am in India, 9:30am in Australia, 10:00am Singapore/Manila/Tokyo/Hong Kong/Taipei, and 10:30am in New Zealand).
-
Commit to at least 2 hours of independent study during the week to embed learnings.
-
Deliver a project presentation (necessary for certification).
​
Desirable:
-
Basic understanding of deep learning concepts.
-
Working knowledge of linear algebra, probability and statistics​.
What can I expect from the application process?
You'll spend about 2-3 hours completing an application form about your technical background and interest in AI safety.
​
While TARA is free for Participants, we use a commitment bond to increase engagement in the program. Here is how it works:
-
You'll put down a deposit when you start, and receive it back once you've attended at least 11 of 14 Saturday sessions and delivered your project presentation.
-
The bond signals serious commitment - both to yourself and to your cohort - and has helped us achieve a high completion rate over our previous rounds.
The bond amounts for Round 2 are:
-
Australian cities: $200 AUD
-
New Zealand cities: $175 NZD
-
Indian cities: ₹3,000 INR
-
Manila: ₱1,800 PHP
-
Singapore: $200 SGD
-
Taipei: NT$2,000
-
Tokyo: ¥12,000 JPY
-
Hong Kong: $800 HKD
Key Dates - 2026
​Round 2 (September - December)
-
Saturday 5 September: City ice-breaker sessions (online)
-
Saturday 12 September: Program starts in-person
-
Saturday 12 December: Final project presentations, followed by celebration meals in each city