Active pilot
AcadAIsist
An agentic evaluation tool built on a human-in-the-loop framework.
Small Language Model Foundation.
Built on lightweight SLMs to make AI-assisted evaluation affordable, efficient, and institutionally deployable
No-Code Faculty Interface
Faculty can enter questions, answer keys, marks, expected length, and rubrics without writing a single line of code.
Student Answer Interface
Students answer directly within the platform, removing the need for uploaded answer sheets or manual formatting.
Semantic Answer Understanding
The tool recognises when the same idea is expressed in different words, helping evaluate meaning rather than memorised phrasing.
Rubric-Grounded Evaluation
Marks are awarded only against faculty-defined rubrics, keeping evaluation aligned with the course expectation.
Faculty-Controlled Knowledge Base
The SLM provides language understanding, while all factual authority comes from the faculty-provided key and rubric.
Human-in-the-Loop Verification
Faculty can review, accept, modify, or reject suggested marks, making the system governable and transparent.
Evidence-Based Scoring
Each suggested score is supported by matched points, missing components, and evaluation justification.
Uncertainty Flagging
Ambiguous or weakly matched answers are flagged for faculty review instead of being blindly graded.
Locally Deployable and Secure
The lightweight architecture allows institutional deployment without depending entirely on external cloud systems.
Progressive Model Refinement
Faculty corrections can be used to improve the system over time and reduce mismatch with human evaluation.
Expandable Academic Assistant
Starting with descriptive answer evaluation, the platform can later grow toward equation handling, note-based question generation, and course support.
Built with
SLMsLangChainPythonReactPostgreSQLpgvector
Roadmap
Q1 · 2026
Closed alpha
Internal rollout with 3 partner labs.
Q2 · 2026
Public pilot
Open cohort of 50 scholars.
Q3 · 2026
Institutional plan
Library-scale ingestion and SSO.
Q4 · 2026
Agentic workflows
Composable review and critique agents.
Q1 · 2027
1.0 release
Stable API, self-hosting, and audit trails.